Robotics cover
Interactive book · Semester 4-8

Robotics

Sweep a robot's work envelope, build an arm from its DH table, solve inverse kinematics both ways, blend a trajectory and steer a mobile robot until its odometry drifts

18 chapters, 22 interactives, few words. Every chapter is mapped to the Anna University, VTU, SPPU, Mumbai University and KTU units it covers: types of automation, robot anatomy and configurations, degrees of freedom and work envelopes, accuracy and repeatability, rotation matrices and homogeneous transforms, Denavit-Hartenberg forward kinematics, inverse kinematics, the Jacobian and singularities, manipulator dynamics, trajectory planning, drives, gearboxes and motor sizing, joint control, grippers, sensors and machine vision, mobile robots, robot programming, and material handling, economics and safety. Preparing for the new GATE 2027 Robotics and Automation (RA) paper? Its Part A.3 robotics topics are chapters 2 to 6, 13 and 14.

  • AUMR3691Robotics
  • AUCME348Modern Robotics
  • VTUBME405DRobotics and Automation
  • SPPU402051CAutomation and Robotics
  • MUMEC604Automation and Artificial Intelligence
  • KTUMET456Robotics and Automation

By Skill-Lync · Updated 17 Nov 2026 · 18 chapters · 22 interactives · free, no sign-up

Chapter 1

Automation and robots: types of automation, definitions, history, laws of robotics and classification

Which syllabus covers thisAU MR3691 Unit I; CME348 Unit IVTU BME405D Module 1SPPU 402051C Units 1, 2MU MEC604 Modules 1, 4KTU MET456 Module 1

What are the types of automation?

Three. Fixed (hard) automation is a machine built for one product, such as a transfer line machining one engine block: very fast and cheap per part, but expensive to build and almost impossible to change. Programmable automation, such as a CNC machine or a robot, can be reprogrammed for a new batch, with some changeover time between batches. Flexible automation is programmable automation with near-zero changeover, so a cell can make a mix of parts one after another. Robots sit in the programmable and flexible groups.

When does automation pay for itself?

Every option has a yearly fixed cost (the machine, spread over its life, plus upkeep) and a cost per part (mostly labour). Cost per part is fixed cost ÷ volume + variable cost, so the expensive option wins only above some volume. Manual work wins for small batches, a robot cell in the middle, and dedicated fixed automation only at millions of parts a year. Move the operator cost and the two machine prices and watch the break-even volumes slide.

Unit cost against production volume: manual work, a robot cell and fixed automation: interactive loads as you scroll

What is the definition of an industrial robot?

ISO 8373:2021 defines it as an automatically controlled, reprogrammable, multipurpose manipulator, programmable in three or more axes, either fixed in place or on a mobile platform, for use in industrial automation. The key words are reprogrammable (new motions without rebuilding it) and three or more axes. A two-axis pick-and-place unit or a single-purpose machine is therefore automation, but not a robot. The word robot comes from Karel Čapek's 1920 play R.U.R., from the Czech robota, forced labour.

Who made the first industrial robot?

George Devol applied in 1954 for a patent on programmed article transfer, granted in 1961, and with Joseph Engelberger built the Unimate. The first one went to General Motors' plant in Ewing Township, New Jersey, in 1961, lifting hot die castings and stacking them. The Stanford Arm (1969) was an early all-electric, computer-controlled arm. Hiroshi Makino's group at the University of Yamanashi built the first SCARA prototype in 1978. Collaborative robots that share space with people arrived in the 2000s.

What are Asimov's three laws of robotics?

Isaac Asimov stated them in his 1942 story Runaround: 1. A robot may not injure a human being or, through inaction, allow a human being to come to harm. 2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law. 3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law. They are fiction, not engineering rules; real robot safety comes from standards such as ISO 10218 (Chapter 18).

How are robots classified?

By arm configuration (Cartesian, cylindrical, polar, SCARA, articulated: Chapter 2); by drive (hydraulic for heavy loads, pneumatic for simple fast moves, electric for most modern robots); by control (non-servo pick-and-place between end stops, servo point-to-point, servo continuous-path); by generation (first: fixed playback; second: sensors and feedback; third: planning and learning); and by use: industrial robots in factories versus service robots, mobile robots, medical, field, underwater and aerial robots.

Predict, then check

A part needs 90 s of manual work at ₹250/h. A robot cell costs ₹25 lakh, a fixed line ₹150 lakh (fixed cost per year = 30% of price). At 50,000 parts a year the cheapest per part is:

Worked examples

Break-even volume: manual vs robot cell · Operator ₹250/h, 90 s per part. Manual: fixed ₹30,000/yr, variable 90/3600 × 250. Robot cell: fixed 0.3 × ₹25 lakh = ₹7,50,000/yr, variable 20% of the labour + ₹0.50
  1. Manual variable cost: 250 × 90/3600 = ₹6.25 per part
  2. Robot variable cost: 0.2 × 6.25 + 0.50 = ₹1.75 per part
  3. Q* = (7,50,000 − 30,000) / (6.25 − 1.75) = 1,60,000 parts/yr

Answer: Above 1,60,000 parts a year the robot cell is cheaper per part

Break-even volume: robot cell vs fixed line · Fixed line ₹150 lakh → ₹45,00,000/yr fixed, variable 5% of the labour + ₹0.20 = ₹0.5125 per part; robot cell as above
  1. Q* = (45,00,000 − 7,50,000) / (1.75 − 0.5125)
  2. = 37,50,000 / 1.2375 = 30,30,303 parts/yr

Answer: About 3.0 million parts a year: below that, the reprogrammable cell is cheaper

Cost per part at 5 lakh parts a year · Same three options, Q = 5,00,000 parts/yr
  1. Manual: 30,000/5,00,000 + 6.25 = ₹6.31
  2. Robot cell: 7,50,000/5,00,000 + 1.75 = ₹3.25
  3. Fixed line: 45,00,000/5,00,000 + 0.5125 = ₹9.51

Answer: Robot cell ₹3.25 per part, about half the manual cost

Chapter 2

Robot anatomy and configurations: Cartesian, cylindrical, polar, SCARA, articulated, joint notation and work envelope

Which syllabus covers thisAU MR3691 Unit I; CME348 Unit IVTU BME405D Module 2SPPU 402051C Unit 2MU MEC604 Module 4KTU MET456 Module 2

What are the main parts of a robot?

The manipulator is a chain of rigid links joined by joints, mounted on a base. The body-and-arm joints position the wrist; the wrist joints orient the tool. The end effector (gripper or tool) does the work. Actuators (electric motors, hydraulic or pneumatic cylinders) drive the joints through gears or belts, sensors report joint positions and the outside world, and the controller runs the program and closes the servo loops. A teach pendant is the hand-held box used to jog the robot and record points.

What types of joints does a robot have?

Two basic kinds: prismatic (sliding, P) and revolute (rotating, R). The classic notation splits these into five: L, a linear joint sliding along its link; O, an orthogonal sliding joint at right angles to its link; R, rotation about an axis perpendicular to the links; T, twisting about an axis along the links; V, revolving, where the output link is perpendicular to the axis. A robot is written body-and-arm : wrist, for example TRR : RT for a jointed arm with a two-axis wrist.

What are the five common robot configurations?

Cartesian (LOO, also gantry or x-y-z): three sliding axes, a box-shaped workspace, simple to program and very rigid. Cylindrical (TLO): a rotating column with vertical and radial slides. Polar or spherical (TRL): a rotating base, a tilting shoulder and a telescoping arm. SCARA (VRO): two horizontal revolute joints plus a vertical slide, stiff vertically and compliant sideways, ideal for fast pick-and-place and insertion. Articulated or jointed arm (TRR): three revolute joints like a human shoulder and elbow, the most common industrial robot.

What is the work envelope of a robot?

The work envelope (work volume) is the space the wrist centre can reach. Its shape follows the configuration: a box for a Cartesian robot, a thick cylinder shell for a cylindrical robot, a slice of a spherical shell for a polar robot, a flat annulus swept up and down for a SCARA, and a near-spherical region with dead zones for an articulated arm. Joint limits cut pieces out of it. Pick a configuration, change the strokes and link lengths, and compare the side section, plan view and volume.

Work envelope of Cartesian, cylindrical, polar, SCARA and articulated robots: interactive loads as you scroll

What does a robot wrist do?

The arm puts the wrist at a point; the wrist turns the tool to the right orientation. A three-axis wrist gives roll, pitch and yaw. When the three wrist axes meet at one point (a spherical wrist), the position and orientation problems separate, which is why most six-axis industrial arms are built that way: the first three joints place the wrist centre, the last three orient the tool. Two-axis wrists are common on SCARAs and palletizers, where the tool only needs to point down and rotate.

Predict, then check

You need to insert pins into holes on a flat PCB as fast as possible, always from above. The best configuration is:

Worked examples

Cylindrical robot work volume · Outer radius 1.2 m, radial stroke 0.7 m (inner radius 0.5 m), vertical stroke 0.8 m, base turns 360°
  1. V = π (R² − r²) h
  2. = π (1.44 − 0.25) × 0.8 = 2.991 m³

Answer: 2.99 m³

SCARA workspace · Link 1 = 350 mm, link 2 = 250 mm, elbow limited to ±150°, vertical stroke 200 mm
  1. Outer radius: 0.35 + 0.25 = 0.60 m (elbow straight)
  2. Inner radius at the elbow limit: √(0.35² + 0.25² + 2·0.35·0.25·cos 150°) = 0.1829 m
  3. Area: π (0.60² − 0.1829²) = 1.0259 m²
  4. Volume: 1.0259 × 0.2 = 0.2052 m³

Answer: An annulus 183-600 mm in radius, 0.205 m³

Polar robot work volume · Outer radius 1.2 m, radial stroke 0.6 m (inner 0.6 m), elevation ±45°, base 360°
  1. V = (2π/3)(R³ − r³)(sin β₂ − sin β₁)
  2. = (2π/3)(1.728 − 0.216)(2 × 0.7071) = 4.478 m³
  3. Check: numerical integration of 2πr² cos(elevation) dr d(elevation) gives the same

Answer: 4.48 m³

Chapter 3

Degrees of freedom, configuration space and workspace

Which syllabus covers thisAU CME348 Unit IVTU BME405D Module 1SPPU 402051C Unit 2MU MEC604 Module 4KTU MET456 Module 2

What is a degree of freedom of a robot?

A degree of freedom (DOF) is one independent way to move. A free rigid body has 6 in space (3 translations, 3 rotations) and 3 in a plane. A robot's DOF is the number of independent joint variables, usually one per joint. To place a tool at any position with any orientation in its workspace, a robot needs at least 6 DOF: that is why the standard industrial arm has six axes. A SCARA (4 DOF) can only point its tool straight down.

How do you find the degrees of freedom of a mechanism?

Count links n (including the fixed ground link) and joints. Start from the freedom of all moving links, then subtract what each joint takes away: in a plane, a revolute or prismatic joint removes 2 of 3 freedoms, a cam or rolling-sliding joint removes 1. In space, each joint removes 6 minus its own freedoms. M = 1 means one motor drives it, M = 0 a structure. Step through the mechanisms, then count the four-bar with the spatial formula and see why the special geometry matters.

Degrees of freedom of a mechanism by the Grübler-Kutzbach count: interactive loads as you scroll

What is configuration space in robotics?

The configuration of a robot is a complete list of its joint values; the configuration space (C-space) is the set of all of them, with one dimension per DOF. For a two-link arm with unlimited revolute joints, each angle lives on a circle, so its C-space is a torus. Motion planners work in C-space: obstacles in the room become forbidden regions of C-space, and a collision-free motion is a path that avoids them. Joint limits cut C-space down further.

What is the difference between task space and workspace?

Task space is the space in which the job is described: for welding, the position and orientation of the torch (6 numbers); for drawing on a whiteboard, just a point on a plane (2 numbers). The workspace is the set of end-effector configurations the robot can actually reach. The reachable workspace is everywhere the tool can get to with some orientation; the smaller dexterous workspace is where it can reach with every orientation.

What is a redundant robot?

A robot with more DOF than its task needs. A 7-axis arm doing a 6-DOF task has one extra: it can swing its elbow while the tool stays still (self-motion), which helps it reach around obstacles, avoid joint limits and singularities. The human arm has 7 DOF from shoulder to wrist for the same reason. The opposite is an under-actuated system with fewer motors than DOF, such as a car (it cannot move sideways) or a passive walking robot.

Predict, then check

A planar linkage has 6 links (including the frame) and 7 revolute joints. Its mobility is:

Worked examples

Four-bar and five-bar linkages · Four-bar: n = 4, four revolute joints. Five-bar: n = 5, five revolute joints
  1. Four-bar: M = 3(4 − 1) − 2 × 4 = 1
  2. Five-bar: M = 3(5 − 1) − 2 × 5 = 2
  3. The five-bar needs two motors; it is the basis of several fast parallel pick-and-place robots

Answer: 1 and 2 DOF

Stewart platform (6-UPS) · Ground + platform + 6 legs of 2 parts each: n = 14. Per leg: universal (2), prismatic (1), spherical (3): j = 18, Σf = 36
  1. M = 6(n − 1 − j) + Σf
  2. = 6(14 − 1 − 18) + 36 = −30 + 36 = 6

Answer: 6 DOF: the six leg actuators set all six coordinates of the platform (flight simulators, machining hexapods)

Why the planar count is needed for planar linkages · Four-bar counted with the spatial formula: n = 4, four revolute joints (f = 1 each)
  1. M = 6(4 − 1 − 4) + 4 = -2
  2. A negative answer predicts an over-constrained structure, yet the four-bar moves
  3. Its four axes are parallel, a special geometry: every link stays in one plane, so only 3 freedoms per link exist and the planar formula applies

Answer: Spatial count -2 (wrong); planar count 1 (right)

Chapter 4

Resolution, accuracy, repeatability and robot specifications

Which syllabus covers thisAU MR3691 Unit IVTU BME405D Module 2SPPU 402051C Unit 2MU MEC604 Module 4KTU not in syllabus

What is the difference between accuracy and repeatability of a robot?

Accuracy is how close the robot gets to a commanded point given in coordinates. Repeatability is how closely it returns to the same taught point again and again. ISO 9283 measures both by sending the robot to the same pose 30 times: accuracy AP is the distance from the mean of the attempts to the command, repeatability RP is the radius around that mean that holds the attempts. Slide the bias and scatter apart to see that a robot can be very repeatable and still inaccurate.

Accuracy and repeatability: 30 repeated moves to one taught point: interactive loads as you scroll

What is control resolution and spatial resolution?

Control resolution is the smallest step the controller can command on an axis: the axis range divided by the number of addressable steps, 2ⁿ for an n-bit position register or encoder. Spatial resolution adds the mechanical errors (backlash, deflection, encoder and gear errors), so it is coarser. A common textbook model treats those errors as a normal spread with standard deviation σ: repeatability ±3σ, accuracy CR/2 + 3σ, spatial resolution CR + 6σ.

What do the specifications on a robot datasheet mean?

Number of axes; payload (the mass it can carry at the wrist, at a stated centre of gravity); reach (from the base axis to the wrist centre); pose repeatability (to ISO 9283); maximum joint speeds; mass; mounting (floor, wall, ceiling); and protection rating (IP class). For example, ABB's datasheet for its small six-axis IRB 120 lists a 3 kg payload, 580 mm reach and 0.01 mm position repeatability. Absolute accuracy is rarely printed, because it depends on calibration.

Why is a robot less accurate than it is repeatable?

Repeatability only needs the same errors every time. Accuracy needs the controller's kinematic model to match the real arm: small errors in link lengths and joint zero offsets, links and gears bending under the payload, gear backlash and thermal growth all shift where the tool really ends up, by a fraction of a millimetre to a few millimetres on a large arm. That is why taught points (lead-through) work well, while offline-programmed points need the robot and cell to be calibrated first.

How is robot accuracy and repeatability tested?

ISO 9283 defines the test: five poses on the diagonals of a cube inside the workspace, each visited 30 times at stated speed and payload, the attained positions measured with a laser tracker or similar instrument. It also defines path accuracy, path repeatability, cornering overshoot and stabilisation time. Because the conditions are fixed, datasheets from different makers can be compared. Measure at the payload and speed you will actually use: repeatability gets worse as both rise.

Predict, then check

A robot returns to within 0.02 mm of the same taught point every time, but lands 1.5 mm from where the CAD coordinates say. It is:

Worked examples

Control resolution and the textbook precision model · 1 m axis, 12-bit position control, mechanical errors with σ = 0.02 mm
  1. CR = 1000 / 2¹² = 1000 / 4096 = 0.244 mm
  2. Repeatability = ±3σ = ±0.06 mm
  3. Accuracy = CR/2 + 3σ = 0.122 + 0.06 = 0.182 mm
  4. Spatial resolution = CR + 6σ = 0.364 mm

Answer: CR 0.244 mm, accuracy 0.18 mm, repeatability ±0.06 mm

What a gearbox does for resolution · 17-bit encoder (131,072 counts per turn), 800 mm from the joint to the tool, encoder on the joint vs on a motor behind a 100:1 gearbox
  1. One count on the joint: 2π / 131,072 = 4.794e-05 rad
  2. At 800 mm: 0.0383 mm per count
  3. Encoder on the motor: the joint turns 100 times less per count → 0.38 µm at the tool

Answer: 0.038 mm direct vs 0.38 µm through the gearbox (backlash, not the encoder, then limits precision)

AP and RP from five attempts · Command (0, 0) mm; attempts (0.10, 0.02), (0.12, −0.01), (0.08, 0.00), (0.11, 0.03), (0.09, 0.01) mm (ISO 9283 uses 30; five keeps the arithmetic short)
  1. Mean: (0.10, 0.01) mm → AP = √(0.10² + 0.01²) = 0.1005 mm
  2. Distances from the mean: 0.0100, 0.0283, 0.0224, 0.0224, 0.0100 mm
  3. l̄ = 0.0186 mm, S = 0.0082 mm
  4. RP = l̄ + 3S = 0.0186 + 0.0246 = 0.0432 mm

Answer: AP = 0.100 mm, RP = 0.043 mm

Chapter 5

Position and orientation: rotation matrices, Euler and roll-pitch-yaw angles, homogeneous transforms

Which syllabus covers thisAU MR3691 Unit II; CME348 Unit IVTU BME405D Module 2SPPU 402051C Unit 5MU not in syllabusKTU MET456 Module 2

What is a rotation matrix in robotics?

A 3 × 3 matrix that describes how one frame is turned relative to another. Its three columns are the unit vectors of the rotated frame's x, y and z axes, written in the reference frame. Because those axes are perpendicular unit vectors, the matrix is orthonormal: its inverse is simply its transpose and its determinant is +1. Multiplying a vector by R re-expresses it from the rotated frame in the reference frame. Robots chain these matrices to know which way every link and the gripper face.

What are the basic rotation matrices about the x, y and z axes?

A rotation by θ about one axis leaves that axis unchanged and turns the other two in their plane by θ. About z, the x axis moves to (cos θ, sin θ, 0) and the y axis to (−sin θ, cos θ, 0), which gives the matrix below. Rotations about x and y have the same pattern with the rows and columns of the fixed axis left as 1 and 0. Every orientation can be built from three of these elementary rotations.

What is the difference between rotating about fixed axes and current axes?

Order matters, and so does which axes you turn about. If each new rotation is about the original (fixed) reference axes, you pre-multiply: the latest rotation goes on the left. If each rotation is about the axes of the frame as it is now (current or moving axes), you post-multiply: the latest rotation goes on the right. The same three rotations can therefore land in two different orientations, which is the most common mistake in exam problems.

What are roll, pitch and yaw angles and Euler angles?

Three numbers that describe an orientation. Roll-pitch-yaw turns the frame by roll φ about x, pitch θ about y and yaw ψ about z, all about the fixed axes, so R = Rz(ψ)Ry(θ)Rx(φ). That is the same matrix as Z-Y-X Euler angles about the moving axes. Robot wrists often use Z-Y-Z Euler angles instead. Every three-angle set has a gimbal lock: at pitch ±90° roll and yaw turn about the same line and one angle is lost. Move the sliders, then switch the order to see the same angles give a different pose.

Rotating a frame by roll, pitch and yaw: the rotation matrix live: interactive loads as you scroll

What is a homogeneous transformation matrix?

A 4 × 4 matrix that packs a rotation R and a translation d into one object, so a point can be moved and turned in a single multiplication. A point p₁ known in frame 1 is found in frame 0 as p₀ = R p₁ + d. Transforms chain by plain multiplication, T₀₂ = T₀₁ T₁₂, which is how a robot computes where its gripper is from link to link. The inverse is not the matrix inverse by brute force: it is [Rᵀ, −Rᵀd; 0, 1].

Predict, then check

A frame turns 90° about x, then 90° about the FIXED z axis. Its own z axis now points along:

Worked examples

Fixed axes vs current axes · Rotate a body 90° about z, then 90° about y. Track the point p = (1, 0, 0) fixed in the body.
  1. About fixed axes: R = Ry(90°) Rz(90°) (the later rotation on the left)
  2. R p: Rz(90°) takes (1, 0, 0) to (0, 1, 0); Ry(90°) leaves it at (0, 1, 0)
  3. About current axes: R = Rz(90°) Ry(90°) (the later rotation on the right)
  4. R p: Ry(90°) takes (1, 0, 0) to (0, 0, −1); Rz(90°) leaves it at (0, 0, −1)

Answer: (0, 1, 0) about fixed axes but (0, 0, −1) about current axes: same rotations, different result

Point through a homogeneous transform, and back · Frame 1 is turned 30° about z₀ and its origin is at d = (2, 1, 0) m. A point is at p₁ = (0.5, 0, 0.2) m in frame 1.
  1. p₀ = R p₁ + d with R = Rz(30°)
  2. x₀ = 0.5 cos 30° + 2 = 2.433 m, y₀ = 0.5 sin 30° + 1 = 1.250 m, z₀ = 0.2 m
  3. Inverse: a base point q₀ = (3, 2, 0) in frame 1 is q₁ = Rᵀ(q₀ − d) = Rᵀ(1, 1, 0)
  4. q₁ = (cos 30° + sin 30°, −sin 30° + cos 30°, 0) = (1.366, 0.366, 0) m

Answer: p₀ = (2.433, 1.25, 0.2) m; q₁ = (1.366, 0.366, 0) m

Roll-pitch-yaw to a rotation matrix and back · Roll φ = 30°, pitch θ = 20°, yaw ψ = 45°, R = Rz(ψ) Ry(θ) Rx(φ)
  1. R₁₁ = cos ψ cos θ = 0.6645; R₃₁ = −sin θ = -0.3420
  2. R₃₂ = cos θ sin φ = 0.4698; R₃₃ = cos θ cos φ = 0.8138
  3. Back: θ = atan2(−R₃₁, √(R₁₁² + R₂₁²)) = 20°, φ = atan2(R₃₂, R₃₃) = 30°, ψ = atan2(R₂₁, R₁₁) = 45°
  4. Single equivalent rotation: cos α = (trace R − 1)/2 gives α = 52.7°

Answer: R₁₁ = 0.664, R₃₁ = -0.342; the angles come back exactly; one 52.7° turn about a single axis gives the same pose

Chapter 6

Forward kinematics and the Denavit-Hartenberg convention

Which syllabus covers thisAU MR3691 Unit II; CME348 Unit IIVTU BME405D Module 2SPPU 402051C Unit 5MU not in syllabusKTU MET456 Module 3

What is forward kinematics of a robot?

Working out where the gripper is, and which way it faces, from the joint values. For a serial arm it always has exactly one answer: attach a frame to every link, write the transform from each frame to the next, and multiply them in order from the base to the tool. The result is the tool's pose as a homogeneous transform. The controller does this hundreds of times a second to display the tool position on the teach pendant.

What are the four Denavit-Hartenberg parameters?

Four numbers per link that fix one frame relative to the previous one. Link length a is the distance between the two joint axes along their common normal. Link twist α is the angle between the two joint axes. Link offset d is the distance along the previous joint axis. Joint angle θ is the rotation about that axis. For a revolute joint θ is the variable and a, α, d are constant; for a prismatic joint d is the variable. Four numbers replace the six a general transform would need.

How do you assign DH frames to a robot arm?

Put each z axis along a joint axis: z₀ along joint 1, z₁ along joint 2, and so on. Put x_i along the common normal from z_{i−1} to z_i (if the axes intersect, perpendicular to both; if they are parallel, any normal, usually the one that makes d zero). The origin goes where x_i meets z_i, and y_i completes a right-handed frame. Then read a, α, d and θ off the drawing row by row. The tool frame sits at the gripper with z pointing out of it.

How do you find the end-effector position of a three-link planar arm?

All three joint axes point out of the page, so every twist α and offset d is zero and each a is the link length. Multiplying the three DH matrices collapses to adding the links as vectors: each link points at the sum of the joint angles before it, and the tool angle is the sum of all three. Drag the joint sliders and watch the DH table, the frames at the end of each link and the tool position update.

Forward kinematics of a three-link arm from its DH table: interactive loads as you scroll

What is the difference between standard and modified DH parameters?

Where frame i sits. In the standard (distal) convention, used in Spong and most Indian syllabi and in this book, frame i is at the far end of link i and the matrix is Rot_z·Trans_z·Trans_x·Rot_x. In Craig's modified (proximal) convention frame i is at the near end, on joint i, and the order becomes Rot_x(α_{i−1})·Trans_x(a_{i−1})·Rot_z(θ_i)·Trans_z(d_i). Both give the same tool pose; the tables differ, so never mix rows from the two.

Predict, then check

A 2R arm has L1 = L2 = 1 m. With θ1 = 90° and θ2 = −90°, the tip is at:

Worked examples

Two-link planar arm · a1 = 0.5 m, a2 = 0.4 m, α = d = 0, θ1 = 30°, θ2 = 45°
  1. DH table: row 1 (0.5, 0, 0, 30°), row 2 (0.4, 0, 0, 45°)
  2. x = 0.5 cos 30° + 0.4 cos 75° = 0.4330 + 0.1035 = 0.5365 m
  3. y = 0.5 sin 30° + 0.4 sin 75° = 0.2500 + 0.3864 = 0.6364 m
  4. Tool angle φ = 30° + 45° = 75°

Answer: Tip at (0.537, 0.636) m, pointing at 75°

SCARA robot (RRP) by DH · a1 = 0.35 m, a2 = 0.25 m, d1 = 0.40 m; θ1 = 30°, θ2 = 60°, quill d3 = 0.15 m
  1. DH rows: (a1, 0°, d1, θ1), (a2, 180°, 0, θ2), (0, 0°, d3, 0). The 180° twist flips z₂ to point down, so the quill extends downwards
  2. T = A1 A2 A3 gives x = a1 cos θ1 + a2 cos(θ1 + θ2), y = a1 sin θ1 + a2 sin(θ1 + θ2), z = d1 − d3
  3. x = 0.35 × 0.866 + 0.25 × cos 90° = 0.3031 m; y = 0.175 + 0.25 = 0.425 m
  4. z = 0.40 − 0.15 = 0.25 m

Answer: Tool at (0.303, 0.425, 0.25) m

Three-link planar arm · a = 0.5, 0.4, 0.15 m; θ1 = 30°, θ2 = 45°, θ3 = −60°
  1. Wrist (end of link 2) as in the 2R example: (0.5365, 0.6364) m
  2. Link 3 points at θ1 + θ2 + θ3 = 15°
  3. x = 0.5365 + 0.15 cos 15° = 0.6814 m; y = 0.6364 + 0.15 sin 15° = 0.6752 m

Answer: Tool at (0.681, 0.675) m, at 15°: set the same angles in the interactive to check

Chapter 7

Inverse kinematics: two- and three-link arms, multiple solutions and reachability

Which syllabus covers thisAU MR3691 Unit II; CME348 Unit IIVTU BME405D Module 2SPPU 402051C Unit 5MU not in syllabusKTU MET456 Module 3

What is inverse kinematics in robotics?

The reverse of forward kinematics: given where the gripper must be and how it must face, find the joint values that put it there. This is what a robot does whenever you command a point in space, a straight-line move or a pick position from a camera. It is harder than forward kinematics because the equations are nonlinear and the answer may not be unique: there can be no solution, one, several or infinitely many.

Why does inverse kinematics have multiple solutions?

Because different arm shapes can put the tool in the same place. A two-link arm reaches most points with the elbow up or the elbow down. A six-axis industrial arm with a spherical wrist usually has eight: shoulder left or right, elbow up or down, wrist flipped or not, and a general six-joint arm can have up to sixteen. The controller picks the one closest to the current posture, or the one the programmer stored with the point, so the arm does not swing through a wild reconfiguration.

What is the workspace of a robot and when is a point unreachable?

The reachable workspace is every point the tool can touch in at least one orientation; the dexterous workspace is the smaller set it can reach in every orientation. For a two-link arm the reachable workspace is a ring between radii |L1 − L2| and L1 + L2. Outside the ring |cos θ2| comes out greater than 1, which is the algebra's way of saying there is no solution. On the outer and inner circles the two solutions merge into one, and joint limits shrink the ring further on a real robot.

How is inverse kinematics solved for a three-link arm or a six-axis robot?

By splitting position from orientation. For a three-link planar arm with the tool angle φ given, step back from the tool along the last link to the wrist point, solve the two-link problem for the wrist, then θ3 = φ − θ1 − θ2. Six-axis arms with a spherical wrist do the same in 3D (Pieper's kinematic decoupling): the first three joints place the wrist centre, the last three set the orientation. Arms without a closed-form solution are solved numerically, by iterating with the Jacobian.

Predict, then check

A two-link arm has L1 = L2 = 0.5 m. How many inverse kinematics solutions does the point (1.2, 0) m have?

Worked examples

Both solutions of a two-link arm · L1 = 0.5 m, L2 = 0.35 m, target (0.55, 0.35) m
  1. cos θ2 = (0.55² + 0.35² − 0.5² − 0.35²)/(2 × 0.5 × 0.35) = 0.1500
  2. Elbow-down: θ2 = +81.37°; θ1 = atan2(0.35, 0.55) − atan2(0.35 sin θ2, 0.5 + 0.35 cos θ2) = 32.47° − 32.06° = 0.41°
  3. Elbow-up: θ2 = −81.37°; θ1 = 32.47° + 32.06° = 64.53°
  4. Check by forward kinematics: both give x = 0.55, y = 0.35

Answer: (θ1, θ2) = (0.41°, 81.37°) or (64.53°, −81.37°)

Three-link arm with the tool angle given · a = 0.5, 0.4, 0.15 m; tool at (0.6814, 0.6752) m pointing at φ = 15°
  1. Wrist: x_w = 0.6814 − 0.15 cos 15° = 0.5365, y_w = 0.6752 − 0.15 sin 15° = 0.6364
  2. Two-link IK for the wrist: cos θ2 = 0.7071, θ2 = ±45.0°
  3. Solution A: θ1 = 30.0°, θ2 = 45.0°, θ3 = 15° − θ1 − θ2 = -60.0°
  4. Solution B: θ1 = 69.73°, θ2 = -45.0°, θ3 = -9.73°

Answer: (30°, 45°, −60°), the angles of the forward kinematics example, or (69.73°, −45°, -9.73°)

Which points can the arm reach? · L1 = 0.6 m, L2 = 0.4 m. Points P (0.9, 0.5), Q (0.1, 0.15), S (0.5, 0.5) m
  1. Reachable ring: |0.6 − 0.4| = 0.2 m to 0.6 + 0.4 = 1.0 m
  2. P: r = √(0.81 + 0.25) = 1.030 m, beyond 1.0 m
  3. Q: r = √(0.01 + 0.0225) = 0.180 m, inside the 0.2 m hole
  4. S: r = 0.707 m, inside the ring: two solutions. A point at exactly r = 1.0 m has one (arm straight)

Answer: Only S is reachable; P is too far, Q too close to the base

Chapter 8

Velocity kinematics, the Jacobian, singularities and static forces

Which syllabus covers thisAU CME348 Unit IIVTU not in syllabusSPPU 402051C Unit 5MU not in syllabusKTU MET456 Module 4

What is the Jacobian matrix in robotics?

The matrix that turns joint speeds into tool speed: v = J q̇. Each column is how fast the tool moves when only that joint turns at 1 rad/s. It changes with the arm's posture, so the same joint speeds give different tool speeds in different poses. For a planar two-link arm it is the 2 × 2 matrix of partial derivatives of x and y with respect to θ1 and θ2. A six-axis arm has a 6 × 6 Jacobian covering linear and angular velocity.

How do velocities propagate from link to link in a manipulator?

Each link's angular velocity is the previous link's plus the new joint's rate about its axis, and each origin's linear velocity is the previous origin's plus the angular velocity crossed with the link vector. Stacking these contributions gives the geometric Jacobian directly: a revolute joint i contributes z_{i−1} × (o_n − o_{i−1}) to linear velocity and z_{i−1} to angular velocity; a prismatic joint contributes z_{i−1} to linear velocity and nothing to angular velocity.

What is a singularity of a robot arm?

A posture where the Jacobian loses rank (its determinant is zero), so the tool cannot move in some direction however the joints turn. For a two-link arm det J = L1 L2 sin θ2: the arm is singular when fully stretched or fully folded. Six-axis arms also have wrist singularities, when axes 4 and 6 line up. Close to a singularity, a slow tool motion demands huge joint speeds, which is why controllers slow down or stop a linear move that passes near one.

How are tip forces related to joint torques?

Through the transpose of the same Jacobian: τ = Jᵀ F. It follows from virtual work, because power in at the joints must equal power out at the tip, τ·q̇ = F·v. A robot uses it to hold a load or to press with a set force without any motion. At a singularity the arm can resist a force along the locked direction with no motor torque at all. Move the arm and the force: the ellipse shows the tip speeds the joints can produce and flattens to a line at a singularity.

Jacobian, manipulability ellipse and joint torques for a force at the tip: interactive loads as you scroll

What is manipulability in robotics?

A measure of how freely the tool can move in a posture. Map every joint-speed vector of unit size through J and you get an ellipse (an ellipsoid in 3D): its semi-axes are the singular values of J. Yoshikawa's manipulability is their product, which for a square Jacobian equals |det J|; it is zero at a singularity. The force ellipse is the reverse: long where the velocity ellipse is short, because the arm can push hardest in the directions it moves slowest.

Predict, then check

A two-link arm is stretched out straight along x (θ2 = 0). Can the tip move straight outwards along x?

Worked examples

Jacobian at a posture · L1 = 0.5 m, L2 = 0.4 m, θ1 = 40°, θ2 = 80° (θ1 + θ2 = 120°)
  1. J₁₁ = −0.5 sin 40° − 0.4 sin 120° = -0.6678; J₁₂ = −0.4 sin 120° = -0.3464
  2. J₂₁ = 0.5 cos 40° + 0.4 cos 120° = 0.1830; J₂₂ = 0.4 cos 120° = -0.2000
  3. det J = J₁₁J₂₂ − J₁₂J₂₁ = 0.1970 m²; check: L1 L2 sin 80° = 0.1970

Answer: det J = 0.1970 m², far from zero: a well-conditioned posture

Joint rates for a tool velocity · Same arm and posture; the tool must move straight up at v = (0, 0.1) m/s
  1. q̇ = J⁻¹ v, with J⁻¹ = (1/det J)[J₂₂, −J₁₂; −J₂₁, J₁₁]
  2. θ̇1 = (J₂₂ × 0 − J₁₂ × 0.1)/det J = 0.1759 rad/s = 10.08°/s
  3. θ̇2 = (−J₂₁ × 0 + J₁₁ × 0.1)/det J = -0.3391 rad/s = -19.43°/s
  4. Near θ2 = 0° det J → 0 and these rates blow up

Answer: θ̇1 = 10.08°/s, θ̇2 = -19.43°/s

Holding a load with the arm stretched out · L1 = 0.5 m, L2 = 0.4 m, arm horizontal (θ1 = θ2 = 0), a 50 N weight hangs at the tip (links weightless)
  1. J at θ = 0: [0, 0; 0.9, 0.4]. The tip must push up with F = (0, 50) N
  2. τ = Jᵀ F: τ1 = 0.9 × 50 = 45 N·m, τ2 = 0.4 × 50 = 20 N·m (the moment arms, as statics says)
  3. A 50 N pull along the arm, F = (50, 0): τ = (0, 0). At this singular posture the structure carries it, not the motors

Answer: τ1 = 45 N·m, τ2 = 20 N·m; zero torque for a force along the arm

Chapter 9

Manipulator dynamics: Lagrange-Euler and Newton-Euler, and the two-link arm

Which syllabus covers thisAU MR3691 Unit III; CME348 Unit IIVTU BME405D Module 2SPPU 402051C Unit 5MU not in syllabusKTU MET456 Module 4

What is manipulator dynamics?

The link between the torques the joint motors apply and the motion that results. Inverse dynamics asks: to follow this trajectory, what torque does each motor need? That is used to size motors and gearboxes and, in computed-torque control, to feed the right torque forward. Forward dynamics asks: given these torques, how does the arm move? That is what a simulator integrates. Kinematics ignores mass and force; dynamics is where they come in.

What is the Lagrange-Euler formulation of robot dynamics?

An energy method. Write the kinetic energy K of every link (translation of its centre of mass plus rotation about it) and the potential energy P, form the Lagrangian L = K − P, and differentiate as below for each joint. No internal reaction forces appear, so it suits hand derivations of two- and three-link arms and gives the equations in a tidy matrix form. The cost is that the number of terms grows quickly with the number of links.

What is the Newton-Euler formulation and how is it different?

A force method applied link by link. An outward pass computes every link's angular velocity, angular acceleration and linear acceleration from the base to the tool; an inward pass applies Newton's F = ma and Euler's moment equation to each link from the tool back to the base, giving the force and torque at every joint. It is recursive and its cost grows only linearly with the number of links, which is why real-time controllers use it. Both methods give the same torques.

What do the inertia, Coriolis, centrifugal and gravity terms mean physically?

Inertia torque accelerates the links and grows with the square of speed for the same path: halve the move time and it quadruples. Centrifugal torque at one joint comes from another link whirling round; Coriolis torque appears only when two joints move together. Gravity torque is there even at rest and is largest with the arm held out horizontally. In slow industrial moves gravity dominates, so controllers compensate it; in fast pick-and-place moves inertia dominates.

Predict, then check

A robot repeats the same move in half the time. The inertia part of each joint torque becomes about:

Worked examples

Holding torque with the arm horizontal · Uniform links m1 = 4 kg, L1 = 0.5 m and m2 = 3 kg, L2 = 0.4 m; payload 2 kg at the tip; arm held horizontal and still
  1. At rest only gravity acts: τ = G(q)
  2. τ1 = g[m1 L1/2 + m2(L1 + L2/2) + mp(L1 + L2)] = 9.81 × (1.0 + 2.1 + 1.8) = 48.07 N·m
  3. τ2 = g[m2 L2/2 + mp L2] = 9.81 × (0.6 + 0.8) = 13.73 N·m
  4. Raise the arm vertical and both fall to zero: gravity torque goes with cos of the link angle

Answer: τ1 = 48.1 N·m at the shoulder, τ2 = 13.7 N·m at the elbow

One-link arm by Lagrange · Uniform rod m = 2 kg, L = 0.6 m pivoted at one end in a vertical plane, θ from the horizontal. Find τ at θ = 30° when θ̈ = 3 rad/s² (and θ̇ = 0)
  1. K = ½ I θ̇² with I = mL²/3 about the pivot; P = mg(L/2) sin θ
  2. Lagrange: τ = I θ̈ + mg(L/2) cos θ
  3. I = 2 × 0.36/3 = 0.24 kg·m²; inertia part 0.72 N·m; gravity part 2 × 9.81 × 0.3 × cos 30° = 5.097 N·m

Answer: τ = 5.82 N·m

The inertia matrix changes with posture · Uniform links m1 = 6 kg, L1 = 0.5 m and m2 = 4 kg, L2 = 0.4 m (the interactive's arm, no payload)
  1. M₁₁ = m1 lc1² + I1 + m2(L1² + lc2² + 2 L1 lc2 cos θ2) + I2, with lc = L/2 and I = mL²/12
  2. Arm straight (θ2 = 0): M₁₁ = 2.5133 kg·m² (same as m1L1²/3 + m2L2²/12 + m2(L1 + L2/2)² by the parallel-axis theorem)
  3. Elbow at 90°: M₁₁ = 1.7133 kg·m²; M₂₂ = 0.2133 kg·m² in both
  4. Shoulder torque to accelerate at 2 rad/s² from rest, elbow locked: 5.03 N·m straight vs 3.43 N·m folded

Answer: M₁₁ falls from 2.51 to 1.71 kg·m², so a folded arm needs 32% less torque to swing

Chapter 10

Trajectory planning: cubic, quintic and LSPB joint profiles, via points and Cartesian paths

Which syllabus covers thisAU MR3691 Unit IV; CME348 Unit IIVTU BME405D Module 4SPPU 402051C Unit 6MU MEC604 Module 4KTU MET456 Module 3

What is the difference between a path and a trajectory in robotics?

A path is only the geometry: the sequence of positions the tool or joints pass through. A trajectory adds timing: where the robot is at every instant, and so its velocity and acceleration. Trajectory planning turns a few taught points into smooth position, velocity and acceleration commands for every joint, sampled every few milliseconds, that respect the motors' speed and torque limits and start and stop without jerks.

How do you plan a cubic polynomial trajectory for a robot joint?

A cubic has four coefficients, enough to meet four conditions: start and end position and start and end velocity. From rest to rest the coefficients reduce to the form below, and the peak speed, at mid-move, is 1.5 times the average speed. With via points the move is split into segments, each a cubic whose end velocity is chosen to match the next segment's start velocity, so the robot flows through the via points without stopping.

Why use a quintic polynomial instead of a cubic?

Because a cubic that starts from rest still jumps straight to its full acceleration at t = 0, a step that means infinite jerk, which shakes the arm and excites vibration. A fifth-order polynomial has six coefficients, so it can also make the acceleration zero at both ends. The price is a slightly higher peak speed (1.875 times average instead of 1.5). Switch between the profiles and compare the acceleration curves at the start and end of the move.

Joint trajectories: cubic, quintic and trapezoidal (LSPB) profiles: interactive loads as you scroll

What is an LSPB or trapezoidal velocity trajectory?

Linear segment with parabolic blends: accelerate at a constant rate for a blend time tb, cruise at constant speed, then decelerate at the same rate. The velocity plot is a trapezoid, which is what most robot and CNC controllers run because it uses the motor's speed and acceleration limits fully. The blend time can be at most half the move; a given acceleration limit fixes it by the formula below, and the move is impossible if the limit is below 4Δ/T².

What is the difference between joint-space and Cartesian-space trajectories?

A joint-space move (MOVJ, PTP) interpolates each joint from its start to its end angle: fast, cheap to compute and never troubled by singularities, but the tool sweeps a curve. A Cartesian move (MOVL, LIN) keeps the tool on a straight line or arc, which welding, gluing and inserting need, by solving inverse kinematics at every step; near a singularity the joints have to speed up sharply. Pick a route and play the move both ways to see the curve and the joint-speed penalty.

Joint interpolation vs straight-line motion between two points: interactive loads as you scroll
Predict, then check

A joint moves 90° rest to rest in 2 s on a cubic. Its peak speed is:

Worked examples

Rest-to-rest cubic · A joint goes from 15° to 75° in 3 s, starting and ending at rest
  1. Δ = 60°: a0 = 15, a1 = 0, a2 = 3 × 60/9 = 20, a3 = −2 × 60/27 = -4.4444
  2. At t = 1.5 s: θ = 15 + 20 × 2.25 − 4.444 × 3.375 = 45°, exactly halfway
  3. Peak speed at t = 1.5 s: θ̇ = 2a2t + 3a3t² = 30 °/s (1.5 × 60/3)
  4. Peak acceleration at t = 0 and t = T: 2a2 = 40 °/s²

Answer: θ(t) = 15 + 20t² − 4.444t³ degrees; peak 30 °/s and 40 °/s²

Cubic into a via point · A segment from 0° to 60° in 2 s starts at rest and must pass the via point at 20 °/s to flow into the next segment
  1. a2 = [3Δ − (2v0 + vf)T]/T², a3 = [−2Δ + (v0 + vf)T]/T³
  2. a2 = (180 − 40)/4 = 35, a3 = (−120 + 40)/8 = -10
  3. Check: θ(2) = 35 × 4 − 10 × 8 = 60°; θ̇(2) = 2 × 35 × 2 − 3 × 10 × 4 = 20 °/s

Answer: θ(t) = 35t² − 10t³ degrees, arriving at 60° moving at 20 °/s

LSPB from an acceleration limit · Δ = 90° in T = 2 s; the joint's acceleration limit is 120 °/s²
  1. Feasible? Minimum acceleration 4Δ/T² = 90 °/s² ≤ 120, yes
  2. tb = 1 − √(120² × 4 − 4 × 120 × 90)/(2 × 120) = 1 − 120/240 = 0.5 s
  3. Cruise speed V = 90/(2 − 0.5) = 60 °/s; check a × tb = 120 × 0.5 = 60
  4. Distance: two blends ½ × 120 × 0.5² × 2 = 30° plus cruise 60 × 1 = 60°, total 90°

Answer: Blend 0.5 s, cruise 60 °/s for 1 s, blend 0.5 s

Chapter 11

Robot drive systems: pneumatic, hydraulic and electric drives, DC and AC servo, stepper and BLDC motors

Which syllabus covers thisAU MR3691 Unit VVTU BME405D Module 2SPPU 402051C Unit 3MU MEC604 Module 4KTU MET456 Module 1

What are the three types of robot drive systems?

Pneumatic, hydraulic and electric. Pneumatic drives use compressed air at about 5-7 bar: cheap, fast and clean, but air compresses, so they are hard to stop anywhere except at a mechanical end stop. That suits simple pick-and-place. Hydraulic drives use oil at around 100-200 bar or more: huge force from a small actuator, used on early industrial robots and heavy machines, but they need a pump, and they leak. Electric drives (servo and stepper motors) are precise, quiet and easy to program, which is why almost every modern industrial robot uses them.

Why do industrial robots use AC servo motors?

An AC servo is a brushless permanent-magnet motor with an encoder on its shaft and a drive that controls its current, speed and position in a closed loop. It has no brushes to wear out, a high torque for its rotor inertia, and it holds position against a disturbance because the drive sees the error and corrects it. Brushed DC servos do the same job in older and smaller machines. A BLDC motor is the same brushless idea driven with simpler six-step commutation from Hall sensors, common in drones and mobile robots.

How does a DC servo motor's speed change with voltage and load?

Torque is proportional to current (T = Kt·i) and the motor generates a back-EMF proportional to speed (Ke·ω). So at a fixed voltage the speed falls in a straight line from the no-load speed V/Ke down to zero at the stall torque Kt·V/R. A heavier load slows the motor; more voltage, or a higher PWM duty cycle, moves the whole line up. The motor settles where its line crosses the load's torque line. Change the duty cycle and the load torque and watch the operating point move.

PWM-driven DC motor: torque-speed line, load line and the operating point: interactive loads as you scroll

When is a stepper motor good enough for a robot?

A stepper moves one fixed step (commonly 1.8°, so 200 steps per revolution) for every pulse, so you can position it by counting pulses without any feedback. That makes it cheap and simple, which is why hobby arms, 3D printers and small gantries use it. The catch: if the load needs more torque than the motor can give, it skips steps and the controller never knows. Steppers also lose torque quickly at high speed. Half-stepping and microstepping make the steps smaller and smoother. Pick a drive mode and send steps to see the rotor follow the coils.

Stepper motor drive sequences: wave, full step and half step: interactive loads as you scroll

How do you select a drive for a robot?

Match the drive to the job. Payload and force: hydraulics for very heavy loads, electric for most loads up to hundreds of kilograms. Precision and programmability: electric servos. Cost and simplicity for two-position motion: pneumatics. Environment: pneumatics and hydraulics with no sparks for explosive areas such as paint shops, electric for clean rooms and food lines. Then check speed, duty cycle, size and weight on the arm, and maintenance. Micro actuators (piezoelectric stacks, shape-memory-alloy wires) cover tiny, precise motions such as micro-grippers.

Predict, then check

A DC motor (R = 1.2 Ω, Kt = Ke = 0.05) runs from 24 V. Its load torque doubles from 0.25 to 0.5 N·m. The speed:

Worked examples

Pneumatic cylinder bore for a gripper · Gripper needs 300 N of push; supply 6 bar (0.6 MPa); cylinder efficiency 85% (seal friction)
  1. F = η·p·πD²/4 → D = √(4F/(η·p·π)) = √(4 × 300/(0.85 × 0.6×10⁶ × π))
  2. D = 27.4 mm
  3. Next standard bore 32 mm gives 0.85 × 0.6×10⁶ × π × 0.032²/4 = 410 N

Answer: 32 mm bore (410 N available, 27.4 mm was the minimum)

DC servo operating point · R = 1.2 Ω, Kt = Ke = 0.05 N·m/A, 24 V, load 0.25 N·m
  1. No-load speed: V/Ke = 24/0.05 = 480 rad/s = 4584 rpm; stall torque Kt·V/R = 1.0 N·m
  2. Current: i = T/Kt = 0.25/0.05 = 5 A
  3. Speed: (24 − 1.2 × 5)/0.05 = 360 rad/s = 3438 rpm
  4. Output power T·ω = 0.25 × 360 = 90 W; input 24 × 5 = 120 W; efficiency 75%

Answer: 3438 rpm, 90 W out, 75% efficient

Stepper step rate and joint resolution · 1.8° stepper, half-stepping, 300 rpm wanted; 5:1 belt to the joint
  1. Half step = 0.9° → 400 steps per revolution
  2. 300 rpm = 5 rev/s → 5 × 400 = 2000 steps/s, one every 0.5 ms
  3. At the joint: 0.9°/5 = 0.18° per step (and the joint turns at 60 rpm)

Answer: 2000 steps/s; 0.18° per step at the joint

Chapter 12

Transmissions and motor sizing: gearboxes (planetary, harmonic, cycloidal), belts, ball screws and choosing the motor

Which syllabus covers thisAU MR3691 Unit VVTU BME405D Module 2SPPU 402051C Unit 3MU MEC604 Module 4KTU MET456 Module 1

Why do robot joints need a gearbox?

Electric motors are small and efficient at high speed and low torque, around 3,000 rpm and a few N·m, while a robot joint needs the opposite: maybe 30 rpm and hundreds of N·m. A reducer with ratio N multiplies torque by N (less friction losses) and divides speed by N. It also divides the load inertia the motor feels by N², which is what lets a small motor control a heavy arm. Robot joints typically use ratios of about 50:1 to 160:1.

How does a harmonic drive work?

It has three parts. An elliptical wave generator on the motor shaft spins inside a thin flexible steel cup with external teeth (the flexspline). The flexspline meshes with a rigid ring with internal teeth (the circular spline), which usually has two teeth more. Each turn of the wave generator walks the flexspline back by two teeth, so one stage gives a large ratio with almost no backlash in a compact, often hollow, package. That is why it sits in the wrist and forearm joints of most arms and cobots. The costs: the flexspline is torsionally springy and efficiency is lower than a planetary gear.

What is the difference between planetary, cycloidal and harmonic reducers?

A planetary gear has a sun, planets on a carrier and a ring; with the ring fixed one stage gives 1 + Zr/Zs, typically 3:1 to 10:1, with high efficiency but some backlash, so higher ratios need two or three stages. A cycloidal reducer rolls a lobed disc inside a ring of pins; a single stage gives a high ratio and very high stiffness and shock capacity, so RV-type cycloidal reducers drive the heavy base, shoulder and elbow axes of large robots. Harmonic drives win on size, weight and zero backlash for the lighter wrist axes.

How do robots turn rotation into straight-line motion?

Ball screws, belts and rack-and-pinion. A ball screw moves its nut one lead per revolution, with balls rolling between screw and nut, so efficiency is about 90% against 20-40% for a plain lead screw; Cartesian and SCARA vertical axes use them. Timing belts are light and fast for long gantry axes and SCARA arms, but they stretch a little. Rack-and-pinion drives travel any length, for gantries many metres long. Harmonic, belt and screw stages are often combined: motor, reducer, then belt or screw.

How do you choose the motor and gear ratio for a robot joint?

Start from the move: angle, time and profile give the joint's peak speed and acceleration. Refer everything to the motor: load inertia ÷ N², load torque ÷ (ηN), motor speed × N. Then check four things: peak torque while accelerating is below the motor's peak torque, rms torque over the whole cycle (including dwell) is below its rated torque, motor speed is below its maximum, and the inertia ratio is reasonable (designers often aim for 10:1 or less for a responsive servo). Change the ratio and the move time and watch which limit you hit first (the three servo sizes use typical round catalogue values, not one maker's data).

Sizing a joint motor and gearbox: ratio, reflected inertia and torque: interactive loads as you scroll
Predict, then check

A gear ratio goes from 50:1 to 100:1. The load inertia felt at the motor becomes:

Worked examples

Harmonic drive ratio and output torque · Flexspline 200 teeth, circular spline 202 teeth (fixed); motor 3,000 rpm and 1.27 N·m; efficiency 75%
  1. N = Zf/(Zc − Zf) = 200/(202 − 200) = 100 (the output turns opposite to the motor)
  2. Output speed 3,000/100 = 30 rpm
  3. Output torque 0.75 × 100 × 1.27 = 95.2 N·m

Answer: 100:1, 30 rpm, about 95 N·m

Ball screw motor torque · Lead 10 mm, axial force 2,000 N, ball-screw efficiency 90%, nut speed 0.5 m/s; compare a 4 mm lead screw at 30%
  1. T = F·l/(2πη) = 2,000 × 0.010/(2π × 0.9) = 3.54 N·m
  2. Speed: 0.5 m/s ÷ 0.010 m = 50 rev/s = 3000 rpm
  3. Lead screw: 2,000 × 0.004/(2π × 0.3) = 4.24 N·m, more torque even with 2.5 times less lead, and it wears

Answer: 3.54 N·m at 3000 rpm

Sizing a horizontal joint · Load inertia 2 kg·m², 90° in 0.4 s (1/3 accelerate, 1/3 cruise, 1/3 brake), dwell 0.4 s; 100:1, η = 0.8, joint friction 5 N·m; servo: Jm 0.26×10⁻⁴ kg·m², peak 3.82, rated 1.27 N·m, 6,000 rpm max
  1. Peak joint speed 1.5θ/t = 5.89 rad/s; acceleration 4.5θ/t² = 44.2 rad/s²
  2. Motor speed 100 × 5.89 × 60/2π = 5,625 rpm (under 6,000)
  3. Accelerating: (0.26×10⁻⁴ + 2/(0.8 × 100²)) × 100 × 44.2 + 5/(0.8 × 100) = 1.28 N·m (under 3.82)
  4. Cruise 0.062 N·m, braking -0.78 N·m, dwell 0 → rms = 0.61 N·m (under 1.27)
  5. Inertia ratio 2/(100² × 0.26×10⁻⁴) = 7.7

Answer: The 400 W class servo fits: peak 1.28 N·m, rms 0.61 N·m, 5,625 rpm, inertia ratio 7.7

Chapter 13

Robot control: point-to-point and continuous path, joint servo loops, PD control with gravity compensation and adaptive control

Which syllabus covers thisAU MR3691 Unit IV; CME348 Unit IIVTU BME405D Module 2SPPU not in syllabusMU MEC604 Module 4KTU not in syllabus

What is the difference between point-to-point and continuous path control?

Point-to-point (PTP) control cares only about where each move starts and ends; the path in between is whatever the joints produce. That is fine for spot welding, machine loading and pick-and-place. Continuous path (CP) control makes the tool follow a specified path at a specified speed, which arc welding, painting, gluing and deburring need. The controller does this by interpolating many closely spaced set-points along the path, one every few milliseconds, and feeding them to the joint servos.

How does a robot joint servo loop work?

Each joint has nested feedback loops in its drive. The innermost loop controls motor current, which sets torque; around it a velocity loop compares the commanded and measured speed; the outer position loop compares the commanded angle with the encoder reading. Inner loops run fastest, so each outer loop sees a well-behaved inner one. Each joint is often controlled independently, treating the rest of the arm's dynamics as a disturbance, which works well with high gear ratios because the reducer divides the load's influence by N².

What does gravity compensation do in robot control?

A PD controller pushes with torque proportional to the angle error and resists with torque proportional to speed. Gravity pulls the link down all the time, so the arm can only hold still where the error torque balances the weight: it sags below the target by roughly G/Kp. Raising Kp shrinks the sag but makes the arm stiffer and more oscillatory; integral action removes it slowly. Gravity compensation adds the model's own estimate of G(q) to the command, so the PD part only has to correct errors. Add payload with compensation off, then switch it on.

PD control of a robot joint, with and without gravity compensation: interactive loads as you scroll

What is computed-torque control?

Computed-torque (inverse-dynamics) control uses the full dynamic model of the arm: the inertia matrix M(q), the Coriolis and centrifugal terms and gravity. It calculates the torque needed for the desired acceleration plus a PD correction, which cancels the arm's nonlinear coupling so every joint behaves like a simple, tunable second-order system. It works only as well as the model: payload, friction and inertia errors leave residual tracking error.

What is adaptive control in robotics?

An adaptive controller changes its own parameters while running, because the robot's dynamics change: the payload changes from one part to the next, and an arm's effective inertia changes several times over as it stretches out or folds in. Gain scheduling switches gains by configuration or payload. Model-reference adaptive control adjusts gains so the arm tracks a reference model's response. Self-tuning control estimates the plant parameters online and redesigns the controller from them.

Predict, then check

A link needs 4.3 N·m to hold it at its 60° target. PD control with Kp = 100 N·m/rad and no gravity compensation. The arm stops:

Worked examples

Sag of a PD-controlled link held horizontal · Uniform link 1.5 kg, 0.5 m; payload 1 kg at the tip; held at 0° (horizontal); Kp = 100 N·m/rad, no gravity compensation
  1. Gravity torque at 0°: (1.5 × 0.25 + 1 × 0.5) × 9.81 = 8.58 N·m
  2. Approximate sag: e = G/Kp = 0.0858 rad = 4.92°
  3. Exact: Kp·e = G·cos e → e = 0.0855 rad = 4.90°
  4. Gravity compensation adds 8.58 N·m to the command and the sag disappears

Answer: About 4.9° below horizontal

Choosing Kd for a joint · Same link and payload in a horizontal plane (no gravity): J = 0.375 kg·m², Kp = 100 N·m/rad, Kd = 8 N·m·s/rad, viscous friction 0.1 N·m·s/rad
  1. J = 1.5 × 0.5²/3 + 1 × 0.5² = 0.375 kg·m²
  2. ωn = √(Kp/J) = 16.33 rad/s
  3. ζ = (Kd + b)/(2√(Kp·J)) = 8.1/12.25 = 0.661
  4. Overshoot = e^(−ζπ/√(1−ζ²)) = 6.3%
  5. Critical damping needs Kd = 2√(Kp·J) − b = 12.15 N·m·s/rad

Answer: 6.3% overshoot; Kd ≈ 12.1 for no overshoot

Set-points for a continuous-path weld · 600 mm straight seam at 100 mm/s; the controller interpolates a new set-point every 4 ms (an assumed interpolation cycle)
  1. Weld time: 600/100 = 6 s
  2. Set-points: 6/0.004 = 1,500
  3. Spacing along the seam: 100 × 0.004 = 0.4 mm

Answer: 1,500 set-points, 0.4 mm apart

Chapter 14

End effectors and grippers: mechanical, vacuum, magnetic and tool end effectors, gripper force analysis and compliance

Which syllabus covers thisAU MR3691 Unit V; CME348 Unit IIIVTU BME405D Module 3SPPU 402051C Unit 4MU MEC604 Module 4KTU MET456 Module 2

What is an end effector in a robot?

The end effector is whatever is bolted to the robot's wrist flange to do the work. Grippers hold parts: for pick-and-place, machine loading and assembly. Tools work on parts: a welding torch or spot-welding gun, a spray gun, a screwdriver, a grinding or deburring spindle, a glue nozzle. The robot itself is general-purpose; the end effector makes it a welder or a packer. A tool changer lets one robot swap end effectors automatically, and ISO 9409-1 standardises the flange bolt pattern so tools fit different robots.

What are the types of robot grippers?

Mechanical grippers close two or three fingers on the part, driven by a pneumatic cylinder, a motor or a linkage; parallel jaws keep the faces parallel, angular jaws pivot. They can grip outside the part or expand inside a bore. Vacuum grippers use suction cups for flat, smooth parts such as sheet metal, glass and boxes. Magnetic grippers (electromagnets or switchable permanent magnets) lift ferrous parts and sheets. There are also needle grippers for fabric, adhesive grippers, and soft grippers with inflatable fingers for fruit and other delicate shapes.

How do you calculate the gripping force of a robot gripper?

With a friction grip, the fingers squeeze with force F and friction μF at each contact face must carry the part's weight plus its inertia force when the arm accelerates. With n contact faces and a safety factor: F = SF·m(g + a)/(μ·n) when the arm accelerates upward, which is the usual worst case. Low friction or high acceleration raises the force fast. Form closure avoids this: if the jaws are shaped around the part (a V or a pocket), geometry carries the load and the squeeze can be small. Change μ and the acceleration direction and watch the grip force move.

Gripper force needed to hold a part while the arm accelerates: interactive loads as you scroll

How do you size a vacuum suction cup?

A suction cup holds with the pressure difference times its area: F = ΔP·πd²/4. A vacuum ejector (a venturi driven by compressed air) or a pump typically pulls 60-80 kPa below atmosphere on smooth, non-porous parts; porous cardboard leaks and holds far less. Divide the load (weight plus inertia force, times a safety factor of at least 2) among the cups, find the area per cup, then round up to a catalogue size. Cups mounted face-down lift best; a cup facing sideways relies on friction and needs a bigger margin.

What is remote centre compliance (RCC) in robotics?

Inserting a peg into a hole with a few hundredths of a millimetre clearance is hard for a stiff robot: a tiny misalignment jams the peg. An RCC device is a passive wrist of elastomer pads or springs that lets the peg shift sideways and tilt about a point near the peg's tip, so contact forces steer it into the hole instead of jamming it. That is passive compliance. Active compliance does the same in software, using a force-torque sensor in the wrist and force control.

Predict, then check

The friction coefficient between the jaws and the part doubles from 0.3 to 0.6 (rubber pads). The grip force needed:

Worked examples

Grip force for an accelerating arm · 5 kg part, two jaw faces, μ = 0.3, arm accelerates upward at 4 m/s², safety factor 2
  1. Load on the jaws: m(g + a) = 5 × (9.81 + 4) = 69.05 N
  2. F = SF × load/(μ·n) = 2 × 69.05/(0.3 × 2) = 230 N
  3. Holding still it would need 2 × 5 × 9.81/0.6 = 164 N: acceleration adds 41%

Answer: About 230 N of squeeze

Vacuum cups for a steel sheet · 2 kg sheet, two cups facing down, lift acceleration 5 m/s², vacuum 60 kPa, safety factor 2
  1. Design load: 2 × 2 × (9.81 + 5) = 59.2 N, so 29.6 N per cup
  2. Area per cup: 29.6/60,000 = 494 mm²
  3. Diameter: √(4A/π) = 25.1 mm → choose 30 mm cups
  4. Check: 2 × 60,000 × π × 0.030²/4 = 85 N ≥ 59.2 N

Answer: Two 30 mm cups (25.1 mm minimum)

Payload includes the gripper · Gripper 2.5 kg, part 5 kg, combined centre of mass 120 mm from the wrist flange; the robot's data sheet allows 12 N·m wrist moment
  1. Wrist moment: (2.5 + 5) × 9.81 × 0.12 = 8.83 N·m (under 12 N·m)
  2. Heaviest part at that offset: 12/(9.81 × 0.12) − 2.5 = 7.69 kg
  3. A lighter gripper, or one that holds the part closer to the flange, raises the usable payload

Answer: OK: 8.8 N·m; parts up to about 7.7 kg

Chapter 15

Sensors and machine vision: encoders, resolvers, proximity, range, touch, force-torque and slip sensors, cameras and image processing

Which syllabus covers thisAU MR3691 Unit VVTU BME405D Module 3SPPU 402051C Unit 4MU not in syllabusKTU MET456 Module 5

What are the types of sensors used in robots?

Internal sensors measure the robot itself: encoders or resolvers for joint position, velocity from the same signals, and motor current for torque. External sensors measure the world. Proximity and range sensors find objects without touching. Touch sensors range from simple limit switches to tactile skins that map pressure. A six-axis force-torque sensor at the wrist, strain gauges on a stiff cross, measures contact forces for assembly and polishing. Slip sensors detect a part starting to slide in the gripper so the grip can tighten. Cameras give the richest information of all.

How does an encoder measure a robot joint angle?

An optical encoder shines light through a slotted disc onto photodetectors. An incremental encoder gives two square waves, A and B, a quarter cycle apart; counting every edge of both gives 4 counts per slot, and which wave leads tells the direction. It only counts from a reference, so the robot must home after power-up unless the count is kept alive by a battery. An absolute encoder reads a unique code for every position, in Gray code so only one bit changes at a time. Resolvers measure angle with rotating transformer windings: rugged, analog and tolerant of heat and vibration. Switch between incremental and absolute and turn the disc.

Optical encoders: incremental A/B quadrature and a 4-bit absolute Gray-code disc: interactive loads as you scroll

How do proximity and range sensors work in robots?

Inductive sensors sense metal a few millimetres away through eddy currents in it. Capacitive sensors sense almost any material, even liquid behind a plastic wall, as a change in capacitance. Photoelectric sensors detect a broken or reflected light beam. Ultrasonic sensors time a sound pulse to the object and back: distance = c·t/2, with c about 343 m/s in air at 20 °C. Laser range sensors either time the light (time of flight, used in lidar) or triangulate the spot's position.

How does a laser triangulation sensor measure distance?

A laser fires a spot onto the target and a camera beside it, offset by a baseline b, sees the spot. The nearer the target, the further the spot's image lands from the centre of the detector: by similar triangles x = f·b/z, so z = f·b/x. Because x shrinks as 1/z, one pixel covers more and more depth far away: resolution gets worse with the square of distance, which is why triangulation sensors are for short ranges and lidars use time of flight. Slide the target away and watch one pixel's worth of depth grow.

Laser triangulation range sensor: range from the spot's position on the detector: interactive loads as you scroll

How does a robot vision system find a part?

A CCD or CMOS camera images the scene under controlled lighting (a backlight that silhouettes the part is the easiest). The image is digitised into pixels, each sampled and quantised to a grey level, 0-255 for 8 bits. Thresholding turns it into a binary image; connected-pixel analysis picks out each blob. The blob's area identifies it, its centroid gives position, and its second moments give orientation; camera calibration converts pixels to millimetres in robot coordinates. Move the threshold until the part is clean and the specks disappear.

Machine vision: threshold an image, find the part and its centroid: interactive loads as you scroll
Predict, then check

A triangulation sensor resolves 0.5 mm at 300 mm range. Moved back to 600 mm, it resolves about:

Worked examples

Encoder resolution at the tool · 2,500-line incremental encoder on the motor, ×4 decoding, 100:1 reducer, tool 1 m from the joint axis
  1. Counts per motor revolution: 4 × 2,500 = 10,000
  2. Joint resolution: 360°/(10,000 × 100) = 0.00036°
  3. At 1 m: 1,000 mm × 0.00036 × π/180 = 0.0063 mm
  4. Real accuracy is far worse: gear backlash, link bending and calibration errors dominate

Answer: 0.00036° per count, about 6.3 µm at the tool

Range from a triangulation sensor · f = 8 mm, baseline 50 mm, 5 µm pixels; the spot lands on pixel 160 from the centre
  1. x = 160 × 0.005 = 0.8 mm
  2. z = f·b/x = 8 × 50/0.8 = 500 mm
  3. One pixel is worth z²p/(fb) = 500² × 0.005/400 = 3.125 mm
  4. At 250 mm the spot is at pixel 320 and one pixel is worth 0.78 mm

Answer: 500 mm, ±1.6 mm from pixel rounding

Centroid and orientation of a blob · Binary blob pixels (x, y), y up: (1,1) (2,1) (3,1) (2,2) (3,2) (4,2) (3,3) (4,3) (5,3); 1 pixel = 0.5 mm
  1. Area 9 px = 2.25 mm²
  2. Centroid: x̄ = 3, ȳ = 2 px → (1.5, 1.0) mm
  3. Central moments: μ20 = 12, μ02 = 6, μ11 = 6
  4. θ = ½·atan2(2 × 6, 12 − 6) = 31.7°

Answer: Centre (1.5, 1.0) mm, long axis at 31.7° to x

Chapter 16

Mobile robots: wheeled, legged, aerial and underwater robots, differential-drive kinematics and odometry

Which syllabus covers thisAU CME348 Unit IVVTU not in syllabusSPPU 402051C Unit 6MU not in syllabusKTU MET456 Module 1

What are the types of mobile robots?

Wheeled robots are the most common: differential drive (two driven wheels and a caster, like a robot vacuum), car-like Ackermann steering, and omnidirectional robots on mecanum or omni wheels. Tracked robots cross rough ground. Legged robots (two, four or six legs) step over obstacles but are harder to balance and control. Aerial robots are mostly multirotor drones. Underwater robots are either tethered remotely operated vehicles (ROVs) or autonomous underwater vehicles (AUVs). In factories and warehouses, AGVs follow fixed routes, while autonomous mobile robots (AMRs) plan their own paths around people.

How does a differential-drive robot turn?

It steers by driving its two wheels at different speeds. Its forward speed is the average of the wheel speeds and its turn rate is their difference divided by the wheelbase. At every instant it rotates about a point on the line through the wheel axle, the instantaneous centre of rotation (ICR). Equal speeds put the ICR at infinity, so the robot drives straight; equal and opposite speeds put it midway between the wheels, so the robot spins on the spot. Change the wheel speeds and watch the circle and the ICR move.

Differential-drive mobile robot: wheel speeds, path and odometry drift: interactive loads as you scroll

What is a non-holonomic constraint in mobile robots?

A wheeled robot on the floor has three pose coordinates, x, y and heading θ, but an ordinary wheel cannot slide sideways. So a differential-drive or car-like robot cannot move straight sideways at any instant even though it can reach any pose eventually, as you do when you parallel-park. That restriction on velocity, not on position, is a non-holonomic constraint, and it makes path planning harder. Mecanum and omni wheels have rollers around their rims that remove it, giving a holonomic robot that can move in any direction.

What is odometry and why does it drift?

Odometry estimates the robot's pose by adding up wheel-encoder distances: each tick of each wheel updates x, y and heading. It needs no external reference, but every error accumulates and never shrinks. A wheel 1-2% larger than the controller assumes, an uncertain wheelbase, wheel slip or a bump all bend the estimated path away from the real one, and heading errors grow into large position errors over distance. Robots therefore correct odometry with an IMU, landmarks, lidar or camera localisation (SLAM). Set the radius error in the interactive above and watch the gap open over 10 s.

How does a quadcopter drone fly and steer?

Four rotors, two spinning clockwise and two anticlockwise so their drag torques cancel. To hover, total thrust equals weight. Climbing needs more thrust on all four. To roll or pitch, the controller speeds up the rotors on one side and slows the other, tilting the thrust so the drone accelerates sideways. To yaw it speeds up one diagonal pair and slows the other, so the reaction torques no longer cancel. A flight controller does this hundreds of times a second from its IMU, since a quadcopter is unstable without it. Underwater robots steer with thrusters and trim their buoyancy instead.

Predict, then check

Left wheel 0.2 m/s, right wheel 0.4 m/s, wheelbase 0.3 m. The robot drives in a circle of radius:

Worked examples

Turning radius of a differential-drive robot · vL = 0.2 m/s, vR = 0.4 m/s, wheelbase 0.3 m
  1. v = (0.4 + 0.2)/2 = 0.3 m/s
  2. ω = (0.4 − 0.2)/0.3 = 0.667 rad/s
  3. R = v/ω = 0.45 m (inner wheel at 0.30 m, outer at 0.60 m)
  4. One full circle takes 2π/ω = 9.42 s

Answer: R = 0.45 m, a full circle every 9.4 s

One odometry update from encoder counts · Wheels 100 mm diameter, 1,000 counts per wheel revolution, wheelbase 0.3 m; in one 0.1 s update the left wheel counts 250 and the right 300; start at (0, 0), heading 0
  1. dL = 250/1,000 × π × 0.1 = 0.0785 m; dR = 300/1,000 × π × 0.1 = 0.0942 m
  2. Distance d = (dL + dR)/2 = 0.0864 m; Δθ = (dR − dL)/L = 0.0524 rad = 3.0°
  3. x = d·cos(Δθ/2) = 0.0864 m, y = d·sin(Δθ/2) = 0.0023 m

Answer: Pose (0.0864 m, 0.0023 m, 3.0°)

Rotor thrust for a quadcopter · 1.5 kg quadcopter; thrust-to-weight ratio 2 for control margin
  1. Hover: each rotor carries 1.5 × 9.81/4 = 3.68 N
  2. Maximum per rotor at thrust-to-weight 2: 7.36 N
  3. In the grams-force that motor data sheets use: 7.36/9.81 × 1,000 = 750 gf

Answer: 3.68 N to hover, about 750 gf peak per motor

Chapter 17

Robot programming: lead-through, teach pendant, motion commands, WAIT/SIGNAL/DELAY, branching, offline programming and ROS

Which syllabus covers thisAU MR3691 Unit IVVTU BME405D Module 4SPPU 402051C Unit 6MU not in syllabusKTU not in syllabus

What are the methods of robot programming?

Lead-through programming: a person moves the robot through the task and the controller records the points, either by driving it with a teach pendant (powered lead-through) or by physically guiding the arm (manual or walk-through, used for spray painting and on collaborative robots). Textual programming: writing the program in the robot's language. Offline programming: building and testing the program on a simulated robot cell on a computer, then downloading it. Most real programs mix them: logic written as text, key points taught.

What is a teach pendant and what are the limits of lead-through programming?

A teach pendant is the hand-held unit with jog keys or a joystick, an enable (dead-man) switch and a screen. The programmer jogs the arm in joint, world or tool coordinates, records points and edits the program. Its limits: the robot is out of production while being taught, accuracy depends on the operator's eye, logic, sensor decisions and calculated positions are awkward to teach, and the program cannot be tested before the cell exists. A robot program is, in the end, a path in space: a list of points plus how to move between them.

What are the motion commands in a robot program?

A joint move (MOVE in the VAL language, MoveJ in ABB's RAPID) drives all joints to the target together: fast, but the tool follows a curve. A straight-line move (MOVES, MoveL) interpolates the tool along a line, needed for welding seams and approaching parts; circular moves follow an arc. Speed, approach and depart distances and the corner blending zone are set per move. Joint and straight-line interpolation of the same two points are compared in Chapter 10.

What do WAIT, SIGNAL and DELAY commands do?

They tie the robot to the rest of the cell. SIGNAL switches an output on or off, for example to close the gripper or tell a machine to start. WAIT pauses until an input reaches a value, such as a part-present sensor or a machine-done signal. DELAY pauses for a fixed time, for example 0.3 s for a gripper to close. Branching (IF ... GOTO, subroutines and interrupts) lets the program react: skip a missing part, reject a bad one or go to a safe position on a fault.

How is a palletizing program written?

Teach only two points: the pick point on the conveyor and the first slot on the pallet. Every other slot is the first slot shifted by whole box pitches, so three nested loops (rows, columns, layers) generate the whole pattern. The robot travels at a safe height above the tallest stack and moves straight down and up to pick and place. Change the pallet size and robot speed, watch the program step through each box and see the cycle time.

A palletizing program: pick, place, offsets and cycle time: interactive loads as you scroll

What is offline robot programming and ROS?

Offline programming tools such as ABB RobotStudio, FANUC ROBOGUIDE, KUKA.Sim and RoboDK simulate the robot and cell, check reach, collisions and cycle time, and post-process the path into the robot's own language (RAPID for ABB, KRL for KUKA, TP and KAREL for FANUC, INFORM for Yaskawa). ROS, the open-source Robot Operating System, is middleware: programs run as nodes that exchange messages, with MoveIt for motion planning. ROS 2 is the current version; ROS 1's last release, Noetic, reached end of life on 31 May 2025.

Predict, then check

A 3 × 2 × 2 pallet takes 57 s at 1,000 mm/s. Doubling the robot speed to 2,000 mm/s makes the pallet time about:

Worked examples

Slot of the tenth box · 3 rows × 2 columns × 2 layers, rows filled first, boxes 300 × 200 × 150 mm, counting boxes from 0
  1. Box 9: layer k = ⌊9/6⌋ = 1, remainder 3
  2. Column j = ⌊3/3⌋ = 1, row i = 3 − 3 = 0
  3. Shift = (0 × 300, 1 × 200, 1 × 150) mm from the first slot

Answer: SHIFT by (0, 200, 150) mm: the first row, second column of the second layer

Cycle time of the first box · Start above the pick point; safe height 400 mm (2 layers × 150 + 100); pick at (−800, 300, 0), first slot at (0, 0, 0); 1,000 mm/s, 0.1 s settle per move, 0.3 s per grip or release
  1. Down 400 mm and up 400 mm at the pick: 2 × (0.4 + 0.1) s; the first MOVE is already there: 0.1 s
  2. Across: √(800² + 300²) = 854.4 mm → 0.954 s
  3. Down and up at the slot: 2 × 0.5 s; grip + release 0.6 s
  4. Total = 3.654 s

Answer: 3.65 s for the first box

Whole pallet and boxes per hour · Same cell, all 12 boxes
  1. Summing all 96 moves and delays: 57.36 s
  2. Average per box: 4.78 s
  3. Boxes per hour: 3600 × 12 / 57.36 = 753

Answer: 57.4 s per pallet, about 753 boxes an hour

Chapter 18

Material handling, automated assembly, robot cells, applications, economics and safety

Which syllabus covers thisAU MR3691 Unit V; CME348 Unit VVTU BME405D Modules 1, 5SPPU 402051C Units 1, 6MU MEC604 Modules 1, 4KTU MET456 Module 1

What are material handling and storage systems in automation?

Material handling moves, stores and tracks parts through a plant. Transport: conveyors, industrial trucks, automated guided vehicles (AGVs, which follow wires, tape or markers) and autonomous mobile robots (AMRs, which navigate by their own maps), monorails and cranes. Storage: automated storage and retrieval systems (AS/RS), carousels and racks. The ten principles of material handling start with planning, standardisation, the unit load and minimum handling, and end with automation and life-cycle cost. An AGV fleet is sized from the deliveries an hour and the time per delivery.

What is automatic identification in material handling?

Reading what a part, tote or pallet is without a person keying it in. Linear barcodes and 2D codes (QR, Data Matrix) are cheap but need a line of sight. RFID tags are read by radio without line of sight, many at once, and can store data; passive tags have no battery and a range of centimetres to a few metres, active tags have a battery and longer range. Machine vision reads printed text and checks the part at the same time. The robot cell uses these reads to pick the right program.

What is an automated assembly system?

A machine that puts parts together without people. Parts arrive loose in vibratory bowl feeders or hoppers, are oriented by selectors and orientors, held back one at a time by escapements and placed by pick-and-place units or robots. Single-station machines do every step at one place; multi-station machines move the work through stations on an in-line transfer, a rotary dial-indexing table or a carousel. Because every station must succeed, the yield of good assemblies is the product of the stations' success rates.

How do you calculate the payback period of an industrial robot?

Simple payback = investment ÷ net yearly saving, where the saving is the labour freed (operators per shift × shifts × yearly cost per operator) and the upkeep (maintenance, spares, energy, programming) is subtracted. A second shift often halves the payback, because the same robot frees twice the labour. Robots are used for welding, painting, palletizing, machine tending, assembly and inspection, and beyond factories in medicine, agriculture, space and underwater work. Change the price, operators and shifts to see when a cell pays back.

Robot cell payback: investment, labour saved and shifts: interactive loads as you scroll

What are the safety standards for industrial robots?

ISO 10218-1 covers the robot itself and ISO 10218-2 the robot application and cell (guards, light curtains, interlocked doors, safe layouts); both were revised in February 2025, the first major update since 2011. A collaborative application, where people and the robot share a space, uses safety-rated monitored stop, hand guiding, speed and separation monitoring, or power and force limiting. Most of the guidance from ISO/TS 15066:2016 on collaborative robots, including force and pressure limits for body regions, now sits in ISO 10218-2:2025.

Predict, then check

A ₹30 lakh robot cell frees 2 operators on each of 2 shifts, at ₹3.5 lakh a year each; upkeep is 8% of the price a year. The simple payback is about:

Worked examples

Robot cell payback and 5-year return · Cell ₹30 lakh; 2 operators freed per shift, 2 shifts, ₹3.5 lakh per operator per year; upkeep 8% of the price per year; no tax or interest
  1. Labour saved: 2 × 2 × 3.5 = ₹14.0 lakh/yr
  2. Net: 14.0 − 0.08 × 30 = ₹11.6 lakh/yr
  3. Payback: 30 / 11.6 = 2.59 years
  4. 5-year return: (5 × 11.6 − 30) / 30 = 93%

Answer: Pays back in 2.6 years; 93% return over 5 years

How many AGVs? · 40 deliveries an hour, 300 m each way at 1 m/s, 0.5 min to load and 0.5 min to unload, vehicles available 85% of the time
  1. Time per delivery: 5 min loaded + 5 min empty + 1 min handling = 11 min
  2. Deliveries per AGV per hour: 60 × 0.85 / 11 = 4.64
  3. N = 40 / 4.64 = 8.63 → round up

Answer: 9 AGVs

Yield of a 10-station assembly machine · Each station succeeds 99% of the time (q = 0.01); target 95% good assemblies
  1. P = 0.99¹⁰ = 0.9044
  2. For 95% overall each station needs 0.95^(1/10) = 0.99488
  3. That is under 0.5% failures per station

Answer: 90.4% good; each station must reach 99.49% for 95%

Keep going

Recommended reading

  • Introduction to Robotics: Mechanics and Control, J. J. Craig
  • Modern Robotics: Mechanics, Planning, and Control, K. M. Lynch and F. C. Park
  • Robot Modeling and Control, M. W. Spong, S. Hutchinson and M. Vidyasagar
  • Industrial Robotics: Technology, Programming and Applications, M. P. Groover, M. Weiss, R. N. Nagel and N. G. Odrey
  • Introduction to Robotics, S. K. Saha
  • Robotics: Control, Sensing, Vision and Intelligence, K. S. Fu, R. C. Gonzalez and C. S. G. Lee
  • Robotics Technology and Flexible Automation, S. R. Deb and S. Deb
  • Introduction to Autonomous Mobile Robots, R. Siegwart, I. R. Nourbakhsh and D. Scaramuzza

How this book was made

Chapters follow the official syllabi listed above. Every interactive's formulas and every worked-example number were checked by script. Explanations, animations and examples are original to Skill-Lync. Topics informed by MIT OpenCourseWare 2.12 (Introduction to Robotics), CC BY-NC-SA. All explanations, animations and examples are original to Skill-Lync; MIT does not endorse this page.

From the exam to a core job

Talk to a counsellor.

Ask how robotics and automation engineers use robotics on real projects, and what the Path Planning & Trajectory Optimization Using C++ & ROS covers.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.