← Previous lessonFree lesson 2 of 3 · lesson 2 of 43Next lesson →

Unlock all 43 lessons of Introduction to ROS/GitHub/LinuxFull course ₹40,000 · paid

Price & syllabus
Free lesson · 36 min · no payment

Self-Driving Software Stack, Localization and Levels of Autonomy

A free lesson from Skill-Lync's Introduction to ROS/GitHub/Linux course, on Grow by Skill-Lync · Last updated 16 Nov 2026 · Session: “Week 1 Session 2”

What will you learn in this free Introduction to ROS/GitHub/Linux lesson?

The lesson outlines the software block diagram: perception, navigation and diagnostic stacks with an emergency stop. It explains lidar time of flight, radar, GPS, IMU and wheel encoders, then localization, obstacle avoidance, pose estimation and path planning, followed by safety eras and autonomy levels L0 to L5.

  • Perception, navigation and diagnostic software stacks
  • Lidar time of flight and radar principles
  • Localization versus obstacle avoidance versus pose estimation
  • Autonomy levels L0 to L5

What comes after lesson 2?

The next free lesson is lesson 3, “What ROS Is: Messaging, TF, Visualization Tools and Drawbacks” (23 min). Later in Skill-Lync's Introduction to ROS/GitHub/Linux course (paid): lesson 4, Week 1 Session 4; lesson 5, Week 1 Session 5; lesson 6, Week 2 Session 1. The full course has 43 lessons (18 h 39 min of video); see the course overview or its price and syllabus.

Lesson transcript

Machine transcript of the lesson audio (lightly cleaned). Technical terms may contain errors; the video is the reference.

Now that we have looked into the key technologies that are available, let's look into some of the software components of these key technologies and how they are applied in the context of autonomous driving. So at first we have this software perception stack which consists of various kinds of sensors like LIDAR, RADAR, GPS and IMU and Wheel Velocity Sensor. All of these sensors feed into the perception part of the software which then determine based on the sensor input data, the localization, obstacle avoidance and pose estimation. This data is again or let's say this calculation of the sensor data is again fed to the navigation stack where we have the top level control and the path planning.

And this navigation part or the navigation stack then determines how and where the vehicle should be controlled using the vehicle interface. So it sends its signal to the steering control or the brake and throttle control. And then this information from both the perception and the navigation stack is sent to diagnostic stack of the software for the autonomous vehicle. For example, it determines the vehicle health status, the data logger, the inter-process control between various components of the car on the autonomous car. And all of this again is sent to a wireless e-stop.

So let's say there is an emergency then the car needs to stop which is done using the wireless e-stop as well as the visualizer GUI or the graphics user interface which is again used for showing the various data that the vehicle is currently calculating and operating on. Now we'll go deep into what is LIDAR. So LIDAR is the full form of LIDAR is light, imaging, detection and ranging. So this is a technology which is commonly used to make high resolution maps in geoinformatics, archaeology and autonomous driving among other examples.

So where was LIDAR first used? So LIDAR was developed under the direction of Malcolm Stitch from the Hughes Aircraft Company and it was eventually used to provide extremely sensitive detection of particular wavelengths for distance objects. Which means that let's say if there is an object which is quite far away from the autonomous vehicle, using LIDAR we can get a very accurate picture of what object is exactly in front of the vehicle. So the principle behind the use of LIDAR is it uses something known as time of flight detection.

So what does time of flight mean is that basically a light source or laser source sends out a pulse of light which then hits an object of interest. And this object of interest then reflects back the light to the detector of the LIDAR system which then calculates the amount of time it took for the light pulse to go from the source of the light to hit the object and come back. And then calculating the difference or the time difference between when the light hits the source or hits the object from the source and the time it takes for it to come back to the detector, it calculates the exact difference or the exact distance of the object from the source of the LIDAR.

So the wavelength for the laser which is used in LIDAR ranges from 10 micrometers to approximately 250 nanometers which is the entire range of infrared to ultraviolet rays. The two kinds of LIDAR which are used mainly for detecting objects are incoherent and coherent. In coherent systems the detection is more sensitive and it allows for a more indirect detection which results in much lower operating power but it needs very complex transceivers. Both the incoherent and coherent employ pulse models that is it sends a micro pulse or a high energy pulse intermittently which means that it does not sense the light out continuously but rather in short bursts of energy when this results in a very high accuracy of LIDAR sensors.

The light that we get from LIDAR sensors are mainly used for pulse based control which gives back us a point cloud of all the data for the object that the LIDAR sensor is detecting. This point cloud data is then processed using a computer to determine what exactly the shape of the object that is being measured by the LIDAR is and this data is again fed to the car or the autonomous driving vehicle. Next we will look into what radar is. So radar was originally developed for ships and maritime navigation vehicles.

So the principle of radar is that instead of light which is being emitted by LIDAR, radar emits radio frequency or radio signals which when these signals come in contact with the object they are usually reflected or scattered in many directions. So let's say if there is a ship or an object in front of the car which needs to be detected by the radar signal. Then the radar sensor sends out a radio signal pulse which then gets reflected from the surface of the object being detected and it comes back to the radar detector and then based on the frequency difference of the signal that was emitted and the signal that was received it determines the shape of the object.

Now what we need to understand about radar is that radar waves scatter in variety of waves depending on the size of the wavelength of the radio waves that is being sent out and the shape of the target. So if the wavelength is much shorter than the target size then the wave will bounce off in a similar way that light is being reflected by a mirror. If the wavelength is much longer than the size of the target the target may not be visible or poorly reflect the radar signal back which will result in bad or poor detection.

So low frequency radar technology is dependent on resonance for detection but not identification of targets. Short radio waves are reflected from curves and corners in a similar way that is piece of glass or reflective objects detect light. So as you might understand from this that radar is not very good for detecting lot of objects and it actually entirely depends on the shape and size of the object which radar is detecting which can result in the quality of the object being detected. Next we come to GPS and IMU so GPS is global positioning system is a way for navigating vehicles and it has been quite in development since the early 1900s.

This is mainly used for determining the position of a vehicle let's say a car or a ship or an airplane using the data from satellites which are in the low earth orbit satellites. And this satellite then determine the exact coordinates of the vehicle based on latitude and longitude and this information is sent down to the GPS device installed in the vehicle. And IMU means inertial measurement units so what this measures is the amount of distance the vehicle is traveling so let's say a vehicle is traveling a certain speed through a certain distance the IMU uses this information to calculate the position of the vehicle.

So this is very similar to Newton's laws of motion calculation. So let's say if a vehicle starts at point A and travels a distance a certain distance with a speed or with a velocity of 10 kilometers per hour and travels for 10 minutes then it means that the vehicle has traveled a distance of 1.6 kilometers. So this thing is calculated by GPS and IMU. Next we have wheel encoders which measure the velocity of the wheel or the rotational velocity of the wheel. So a sensor is placed on top let's say a top of the wheel part and when the wheel makes one complete rotation then it determines that one complete rotation has been achieved and based on the diameter of the wheel it determines the actual speed of the vehicle.

So all of this information is again sent to the localization part of the software. So what does localization mean. So localization means let's say you are standing in front of a statue in a marketplace in your locality. And your friend has called you to ask you your current whereabouts so that he can come to meet you. So what do you do exactly to give your friend the location of your presence so that he can find you easily. ** ** ** ** ** You tell him that you are located 5 feet away from occurrence from the statue that you are standing under, which is exactly opposite to a certain shop with a certain name.

And based on this data, your friend can triangulate your position accurately and come and meet you. So similarly, a vehicle also needs to detect certain landmarks in order to determine the exact location of the vehicle. And this is done using GPS, for example, and the IMU data and the wheel velocity, because all of these things actually help the vehicle triangulate its original location. So why cannot we use LIDAR or RADAR to do this localization? Because as you might understand, lots of roads and lots of neighborhood might have a very similar geometrical characteristics.

So let's say street A might have similar looking four buildings, and street B might have the same shape of buildings, even though it is a different color. But the data or the point cloud data that we will get from the LIDAR and the RADAR will be just geometrical rectangles, which will unfortunately make completely impossible for determining the location of the vehicle. So we need to know the exact latitude and longitude of the vehicle, with which we can actually calculate the location of the vehicle. And this data is given to us by the GPS.

Now, a GPS has a certain amount of inaccuracy. Let's say a GPS has a tolerance of plus minus 5 meters. And therefore, this is the main reason why GPS cannot be used for indoor autonomous driving. It can be only used in external environment, where the accuracy is not so much important. So, as you might understand, GPS is not so accurate. Therefore, we also need an additional sensor to determine the exact location of the vehicle. So exact location of the vehicle becomes necessary when the vehicle needs to avoid collision or let's say stop in front of a signal post.

So, in that scenario, if the vehicle oversteps the line by 5 meters, that can be quite fatal because then it will mean it has crossed over to another road and it can collide with any other vehicles coming from another side. So that is why it's very important to combine data from GPS and IMU to get the accurate position of the vehicle. Now that we have explained localization, let's go into obstacle avoidance. So, for obstacle avoidance, it's not really necessary to know the exact location of the vehicle because the vehicle can encounter an obstacle in any scenario.

And what we need to exactly know in this case is the exact distance between the vehicle and the obstacle. This data is usually best received by using sensors such as LIDAR and RADAR. Because I mentioned that a GPS has an inaccuracy of plus minus 5 meters, so as you might understand, for obstacle avoidance, plus minus 5 meter is too much of a tolerance to be allowed. Whereas, LIDAR and RADAR provide accuracies in the range of millimeter, which is actually very useful for helping us in avoiding obstacles.

Next, we come to the case of pose estimation. So, pose estimation becomes very important when you want the vehicle to move to a certain location. So, let's say if we want the vehicle to move from location A to location B, then we also need to know the location of the vehicle along the way from A to B because it might so happen. Let's say there is a diversion in the road due to road closure or accident or road repair work. Then we need to know the location of the vehicle using localization, but we also need to estimate the pose or the pose of the vehicle related to the goal destination.

Because if we do not do that, it might so happen that the vehicle, while driving from pose A to pose B, and when it encounters a certain diversion, it measures or it calculates a long way around the obstacle, which actually is very unnecessary. So, there might be some short way to reach the goal point or point B even after avoiding the obstacle, but if we do not know the pose of the vehicle, we won't be able to determine the exact or the minimum route that the vehicle can take in order to reach goal point or point B in the shortest amount of time.

This data from the perception stack is again fed to the navigation stack, where we actually do top level control. So, what do I mean by top level control? So, in top level control, as you might have understood that it is just a way for the driver to steer the car using the steering wheel. And this information needs to be given to an autonomous vehicle as well to determine in which direction the vehicle should move. Then you might ask, what is path planning necessary for? So, path planning can be useful, let's say, if you are not only going from point A to point B, but you need to stop at three other locations.

So, let's say you are going to a picnic with your friends and while driving to the picnic spot, you need to pick up three of your friends. So, in this case, you need to plan the shortest way from your house to your first friend's house, then from your first friend's house to your second friend's house, to your third friend's house and finally to the picnic spot. In order to find out the best possible route that you can take, the shortest possible route that you can take, you need to do path planning.

We humans are very good at determining the path. We just look at the GPS data and we can find out the shortest path. But for an autonomous vehicle, it needs to do all of this calculation by itself. So, here in this case, path planning becomes very necessary. And lastly, all of this data needs to be given to the vehicle interface, that is, it needs this data after it has determined how to reach its destination, how to avoid obstacles, its current location, the vehicle needs to actually move, right?

So, this data is given to the steering control, which steers the car, and it's also given to the throttle or the accelerator, as well as the brake, and so on. So, just because a car is moving, it's sensing the environment and it's moving and it's determining the path and making certain decisions to avoid obstacles, it does not mean that the vehicle will be doing all of these things correctly at this time. So, it helps us immensely, the human driver or at least the rider of the car, to know the current condition of the autonomous vehicle.

So, that is something which is taken care of by the vehicle health status or the data logger on the inter-process control. In the vehicle health status, we come to know the current status of the vehicle. So, for example, what temperature the engine has, whether the tires have enough air pressure or not, whether the computations done by the vehicle is correct, or whether the data that we are receiving from the sensors is correct. So, all of this information is given by the vehicle health status block. Next, we also need to store all of this data and the errors.

Let's say if a vehicle encounters an error and this error happens quite frequently with one car, we need to know exactly when and why this error happens so that we can fix this error. And because it's an autonomous driving vehicle, which actually runs on our software, if we fix the error from one vehicle, we can actually transfer this error or this bug fix to all the other vehicles by just modifying our software. So, this is a great advantage of autonomous vehicles. So, for that purpose, we store all our data in the data logger.

Next, we also need to have inter-process control. What does inter-process control mean? It means that since there are so many components of the vehicle, so let's say the perception stack, the navigation stack, and so on, the communication between these stacks also needs to be properly done. So, let's say even if the sensors like LIDAR and RIDAR are getting the proper data, it does not mean that they are sending the proper data to the navigation. It might so happen that the Ethernet wire or the TCP IP connection to the navigation stack has developed certain kind of error or there is some bugs or some noise interference in the wire itself.

So, this is actually degrading the quality of the data that we are getting from the sensors. So, the quality of the communication between the various blocks also needs to be maintained and checked properly from time to time in order to ensure that they are working correctly. So, this is where the inter-process control comes in. Inter-process control takes into account as well as the communication between the steering control and, let's say, the actual movement of the vehicle and the navigation, communication between the navigation stack and so on.

Now, as with any machine, it might so happen that even if all the components of the software, the perception stack, the navigation stack, and the vehicle interface works perfectly, there might be an emergency due to some external environmental conditions or some other conditions which are actually not in our hand. So, in that case, we need to ensure that the vehicle stops and it does not cause A very big damage. So for that we need this wireless emergency stop button or it can be a mechanical emergency stop button as well.

And all of this data is then visualized using Visualizer GUI so that the user or the driver of the car or in case of a fully autonomous driving vehicle, the passengers of the car have enough information about the current state of the car and the condition of the car. So now that we have received a general overview of the software block diagram, let's see what is the five eras of safety mean. The five eras are defined for safety with the development of self-driving vehicles. And with each era we have advanced in the safety features.

So initially in 1950s to 200s safety was mainly about seat belts, anti-lock brakes, cruise control and so on which were actually the mechanical safety features. So this safety features ensured that in case of a crash, when a vehicle crashed with another vehicle, the idea was to safeguard the health and the condition of the passengers and the drivers. So for that we had seat belts, which actually statistically showed that it reduces the number of fatal injuries in case of a vehicle accident. We also developed anti-lock braking systems and this provided us immense control of the vehicle when the roads have slippery conditions due to snow or due to heavy rain.

In these conditions it was seen that when the brake of the vehicle was pressed quite strongly, the vehicle still skidded along and the brakes did not work properly because the tires were slipping due to the surface conditions. So for that anti-lock braking systems were developed. From 2000 to 2010, we started seeing advanced safety features which were based on electronics. So let's say we had electronic stability which ensured that even if the road condition is bad, that the car control was quite stable enough and the car did not just move around from left to right in random ways.

We developed systems for developing blind spot detection. So as you might know, if you have driven a car, that even with rear view mirrors, there are some places where if there is a car coming from behind, we are not able to see the car using just the rear view mirrors. So for that cases, we developed some extra convex mirrors which actually helped us look into the blind spot. Along with that, we also developed small camera systems which were actually integrated in the car which helped us again look into these blind spots.

Along with that, from 2010 to 2016, we developed advanced driver assistance features which was actually the L1, L2 autonomous driving system which ensured that when we are driving in a long stretch of road, let's say on a highway or straight roads, we can allow the car to drive at a certain speed and the driver then need not interfere for every driving condition of the car. That is the car can maintain its own acceleration and if it detects an obstacle, it can brake and slow down the speed and notify the driver as well.

Then we also developed lane assistance systems where let's say the vehicle when it needed to change lane, it could do it autonomously. So these things were developed under the system or under the program of advanced driver assistance feature. Then we developed this review video system, we had automatic emergency braking. This is most important safety feature that we have developed because it ensured that in case of an emergency, let's say the driver has failed to activate the brake in proper time. Then the vehicle activates the brake on its own using based on automation and this actually helped save a lot of lives.

This happens quite often as you might imagine because let's say you are on a highway and you are driving at a certain speed and the car in front of you is also driving but due to some reason the brake light of the car driving in front of you doesn't turn on even though it brakes and the speed of the car suddenly reduces. And if you fail to detect that and you do not brake your car in proper time, you will crash into the car traveling in front of you.

So this is where the autonomous emergency braking system is of great help. Then we had similar to this autonomous emergency braking, we extended it to include pedestrian autonomous emergency braking. So as you might understand from the name, if the vehicle detects a pedestrian which appears suddenly from blind spot or from certain blocking view, then the vehicle activates this pedestrian autonomous emergency braking to ensure that the car stops and the pedestrian is not hit by the car. Rear automatic emergency braking is similar to the other automatic emergency braking systems as well where let's say you are driving the car in reverse gear and you want to park in a certain tight spot and you cannot see some blind spot.

And if it's dark and if it's dark, you might not see quite well as well in case if you are near to an obstacle, then the vehicle will activate the emergency braking to ensure that you do not hit that obstacle and damage your car. Then we also had this rear traffic alert and lane centering assist so lane centering assist actually helps the driver to drive the car along the middle of the lane so that he doesn't rear off into any of the side lanes. And from 2016 onwards, we started moving towards L3 and L4 autonomy using partially automated safety features, lane keeping assist where the car can determine itself what lane it is driving in and ensure that the car always travels within the lines of these lanes.

We had adaptive cruise control where you can set the car to not only drive with a certain speed but also stop in case an emergency obstacle comes up. Then we had this traffic jam assist which ensure that the car will travel at a very slow speed when it encounters traffic jam. And lastly the self park this self park or car call where you just call out the car to come to you from a parking space has been developed by Tesla and you can see a certain video where you with just a press of a button.

The car just comes out from a parked place to the place where the driver is standing. This feature is called summon and from 2025 onwards it has been planned to have full automated safety features and highway autopilot where the car can safely drive from one place to another using a highway intersection and all kinds of navigating all kinds of city traffic and so on. Levels of autonomy. So what are levels of autonomy. So what are levels of autonomy. There are actually six levels of autonomy and from the order of 0 to 5.

So I'll just go briefly into the details of each level so that you can understand what each stand for. So L0 means that the vehicle features has no automation that the driver takes care of everything from steering acceleration basic control breaking and so on. So the entire perception is dependent on the driver based on what the driver sees and hears and then the driver has to take decision based on what he perceives. Then when if the fallback system that we have increase the automation phase is the driver itself because there is no automation involved in the vehicle.

The there is no other fallback mechanism. The driver is the one and all who is controlling the car based on his perception. Next we come to L1 automation where the driver assistance is features are used. In this case where like we discussed in the previous slides that there is basic safety features like lane track and so on which can helps the driver to keep the car in the lane and so on. In L2 automation where we had this cruise control system where the driver can set the vehicle to have a certain speed and the vehicle will continuously travel in that speed until it encounters a certain obstacle where it will just stop.

So the environmental perception is still done by the driver and the fallback when the automation phase is also the driver himself. Then we have conditional automation where we can set the speed of the vehicle and we are using cruise control mode where we have certain we have a technology called adaptive cruise control where the vehicle can travel at a certain speed on the highway. And if the vehicle traveling in front of the vehicle slows down or speeds up, then the vehicle can actually adapt to that speed and increase and decrease the speed of the car and so on.

Here the perception is done entirely by the vehicle itself as you might imagine using various kinds of sensors like lidar and radar and so on and in case this fails the driver has to intervene and take control of the vehicle. Now high automation is L4 autonomy where the vehicle is completely capable of driving from one place to another. You can play the time of drive from another. So you play the next phase. So that we have a hit zatveler then That we have any other headphones.

The tune is extremely bolt in this technology. Navigating the streets, looking into the traffic signal and ensuring it follows all the road signs and so on. This can be used mainly for traveling on the highways and so on and not exactly in high density city traffic where the vehicle might make some errors and might stop. And lastly, we have L5 or full automation where the vehicle can navigate any kind of traffic from city traffic, from dense traffic, from random chaotic scenarios where the driver need not intervene at all.

Here, the perception is again entirely done by the vehicle and the sensors on board. Even if the vehicle encounters some problems, the vehicle itself should be able to come out of the situation by itself without any human intervention. So lastly, levels of autonomy in the brief, this is how it looks like. So level 1 is assistant driving where you have to place your hands always on the steering wheel and you have to steer the car and the vehicle will just assist you and show you what is the best way to drive a car, follow lanes, slow down the speed and so on.

In level 2 autonomous driving, you can keep your hands off the steering wheel where the vehicle will mostly travel on its own accord, like it will determine the speed of the vehicle in front of it and speed of the vehicle behind it and so on. And in case an emergency happens or in case the car is unable to navigate a scenario properly, then you as the driver need to step in and ensure that the scenario is properly handled. In level 3 autonomous vehicles, you can just allow the car to drive itself.

You need not perceive or need not do the perception task for the vehicle. And only in case of extreme emergency, where let's say a very dense traffic scenario happens or where there are lack of traffic signals and the road condition is bad, then you need to step in. Otherwise, for highways and normal city roads, the vehicle can drive itself completely. In level 4 autonomous systems, you do not need to do anything. You can just be inside the car and even the most emergency scenarios can be handled by the car itself, like it can navigate high density traffic and so on by itself.

And it can just, even if it encounters some big problems, it can solve by itself. This is what happens in level 4 autonomy. In level 5 autonomy, then the driver need not pay any attention at all. The car will be able to determine all the scenarios by itself completely and navigate from point A to point B. So, let's look at the current market. In the current market, the following companies are actually the leaders in autonomous driving and they have made quite a bit of progress in the level of research and the current production issues and in the current implementation of autonomous driving technology in the road itself.

Among these companies, Daimler, General Motors, Google, Tesla, Toyota, Volkswagen, Uber and Lyft, Tesla is the leading company which has already implemented a lot of autonomy features in all of its vehicles. And because Tesla has a large number of vehicles on the road, it is continuously collecting a lot of data from all of its vehicles and this data is actually helping it improve its autonomous driving system on a day-to-day basis. Google, unlike Tesla, has created some autonomous cars using its OMO subsidiary company and these cars are not actually deployed on the road, they are not commercial vehicles, they are just used for research and development purposes where these vehicles drive around certain roads and collect data which is then again used to make the autonomous driving system of Google better and better.

Apart from these two leaders of the market, German car makers like Daimler and Volkswagen have also invested quite a lot in autonomous driving systems and they are currently developing, trying to develop their own autonomous driving systems. But unfortunately, they have not deployed any of their technologies currently on the road and along with that, Uber and Lyft, they are not exactly car companies, they are cab companies and for them, this technology will be a game changer for them because then they can have vehicles without drivers and they can just own these vehicles completely.

And these vehicles can operate throughout the day 24-7 without any human intervention and they can just generate huge amount of money for the companies. So, as you might imagine, this autonomous driving technology is actually a huge opportunity for companies which are into traveling and logistics because if you develop this technology for your car fleet or your truck fleet, you can remove the human drivers entirely from your operation and you can save a lot of money in that that you do not need to pay your drivers and also in the step that you can actually have your cars and trucks operate 24-7 throughout the night without stopping.

So, that is the reason why all of these companies are deeply investing into the self-driving technology.

Want the rest of Introduction to ROS/GitHub/Linux?

Get the full course syllabus and fees.

The full course is paid (₹40,000). We'll send the syllabus and call you back within one working day.

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.

Full course · paid

Unlock the full Introduction to ROS/GitHub/Linux course

The remaining 40 lessons, mentor-reviewed projects and a certificate come with the paid course (₹40,000). Leave your details: we'll send the syllabus and fees and call you back.

Get the syllabus and fees

A counsellor calls you back within one working day.

By submitting you agree to our Terms and Privacy Policy.

Grow by Skill-Lync · free lesson

Start watching

Free, no payment. Leave your details once and every free lesson on Grow plays straight away. A Skill-Lync counsellor may call you about the full Introduction to ROS/GitHub/Linux course.

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.