Fullscreen HTML5Med Res (??MB)HTML4

Autonomous Vehicles ADVANCED DRIVER ASSISTANCE SYSTEMS Driving automation includes fully autonomous vehicles (that can drive entirely by themselves) as well as advanced driver assistance systems (ADASs), which make a human driver’s job easier. Both technologies have made significant strides in recent years. By Dr David Maddison, VK3DSM The ‘future of driverless cars’ from an advertisement in the Philadelphia Saturday Evening Post, 1956. Source: www.saturdayeveningpost.com/2018/05/driverless-cars-flat-tvs-predictions-automated-future-1956/ 12 Silicon Chip Australia's electronics magazine siliconchip.com.au T his article will be about automation in ground vehicles only; we have previously discussed aerial automation in several articles, including the recent one on Drones (also known as UAVs) in the September issue. We also discussed autonomous underwater vehicles in the September 2015 issue, and autonomous agricultural vehicles in June 2018. Classifications Whether or not a vehicle is autonomous is not a simple yes/no answer; there are different levels of autonomy. Thus, there are several schemes to categorise levels of vehicle automation. One of the most commonly used is from the Society of Automotive Engineers (SAE), embodied in their J3016 standard. It defines six levels of vehicle automation. For SAE levels 0-2, the driver is fully driving the vehicle and remains in complete control. Level 0: No driving automation. The vehicle may provide warnings and momentary assistance only, such as automatic emergency braking, blind spot warning and lane departure warning. Most entry-level cars on the market today are at this level. Level 1: Partial automation with a single feature for the vehicle to control, like steering, braking or acceleration. May include lane centring or adaptive cruise control. Many cars on the road today have one of these features. Level 2: Partial driving automation. The vehicle can control (when necessary) steering, braking and acceleration, such as lane centring and adaptive cruise control. A reasonable proportion of cars on the road today have both of these features, and they come on most new higher-end vehicles. Level 3: Conditional driving automation. This includes environment detection, with capabilities like automated overtaking or negotiating traffic jams. The driver must be prepared to take control of the vehicle when required. Examples include Audi A8L Traffic Jam Pilot, Mercedes Benz Drive Pilot, Honda Legend Traffic Jam Assist and BMW Personal Pilot L3. Note that these systems may not be approved in certain locations. For SAE levels 4-5, the driver is not usually required to take control of the vehicle (and may not be able to, as it might not have controls). siliconchip.com.au Level 4: High level of driving automation. The vehicle drives itself under nearly all circumstances. An example is a driverless taxi for use on local roads (eg, Waymo’s One; https:// waymo.com/waymo-one/), shuttle buses in controlled urban environments, delivery services with trucks (Gatik is partnering with Isuzu) and public transportation. Mercedes Benz is trialling level 4 driving on various roads in Beijing. Pedals and steering wheel may not be fitted to a Level 4 vehicle. Such a vehicle likely cannot go off road. Level 5: This is similar to level 4, but more advanced. Whereas level 4 is fully automated, it is restricted to certain structured environments, like road networks. Level 5 has full driving automation under all possible circumstances, including off road. There is currently no example of a widely available car that meets the level 5 criteria. How do autonomous vehicles work? An autonomous vehicle requires many integrated systems to function. That includes multiple sensors to sense and map the environment; actuators to operate systems like steering or brakes; algorithms to guide tasks like parking, lane keeping, or collision avoidance; machine learning to handle a range of scenarios; powerful computers to orchestrate this all; and complex software running on reliable operating systems. A multitude of data from the sensors must be brought together in a process called sensor fusion. Sensor fusion involves merging data from numerous sensors to create a more comprehensive and accurate view of the environment than can be supplied by individual sensors. It is equivalent to how a human combines information from multiple senses (sight, hearing, balance etc). The controlling computer receives instructions from a person about where to go, then plots a route and sends appropriate instructions to the actuators to move the vehicle in the required direction. At the same time, the vehicle is constantly monitoring its environment for collision avoidance, lane keeping, observing speed limits, stopping at signals and stop signs, and observing other traffic rules. Australia's electronics magazine Important Developments Some significant developments toward advanced driver assistance systems and autonomous vehicles are: 1939 the GM Futurama display at New York World’s Fair prophesied a future in which there were semi-automated vehicles equipped with lane centring, lane change & blind spot assist systems, as described in the book Magic Motorways by Norman Bel Geddes. 1952 GM introduced the Autronic Eye, an automatic headlight dimming system, on some Oldsmobile and Cadillac models. 1958 Chrysler offered cruise control, invented by a blind engineer, Ralph Teetor. 1964 Twilight Sentinel was introduced on some Cadillac models, controlled by a photocell to sense ambient light levels and turn the headlights on or off. It was introduced in other models throughout the 1970s and later. Some versions switched on the lights whenever the wipers were activated, to improve safety in low-visibility conditions. 1977 Japan’s Tsukuba Mechanical Engineering Laboratory developed an experimental car that could drive itself on specially marked streets. 1978 Rain sensing wipers were invented by Australian Raymond J. Noack (siliconchip.au/ link/ac6o). 1989 the Volkswagen Futura concept car had four-wheel steering to autonomously manoeuvre into parking spots. 1992 the Mitsubishi Debonair used lidar to warn the driver if they were too close to the vehicle ahead, but couldn’t control the vehicle. 1995 the Mitsubishi Diamante had an adaptive cruise control using lidar but could not apply the brakes. 2003 Honda introduced the Collision Mitigation Brake System to automatically apply the brakes if it detected a collision was imminent. 2004 DARPA held their inaugural Grand Challenge, a series of competitions to encourage the development of “autonomous ground vehicles capable of completing a substantial off-road course within a limited time”. 2006 the Lexus LS460 was sold with a Lane Keep Assist feature that steers vehicle back into lane if it deviates. 2015 Tesla offers the “Autopilot” feature on their Model S. 2019 Mercedes Benz and Bosch test automated valet parking at Stuttgart Airport in Germany, to guide a car autonomously to a pre-booked parking place. 2023 Mercedes Benz’s DRIVE PILOT system is approved in Nevada, USA, to drive on certain freeways during daylight below 40 miles per hour (64km/h). 2024 BMW obtained approval for Personal Pilot L3 in Germany, similar to Mercedes DRIVE PILOT. October 2025  13 Figs.1 & 2: the architecture of a typical autonomous vehicle. ML = machine learning, AI = artificial intelligence, DL = deep learning, UI/UX = user interface/user experience, AUTOSAR = Automotive Open System Architecture, ROS = robot operating system, RTOS = real time operating system, V2X = vehicle to everything. It will also monitor itself, to ensure sufficient fuel or battery charge, while looking for places along the way to refill. Figs.1 & 2 show the generic hardware and software architecture of a typical autonomous vehicle, along with information flows and actions. Environment sensing Autonomous vehicles, or vehicles with ADAS (Advanced Driver Assistance System), need ‘eyes’ to see the environment around them, as well as other sensors. The main sensors are lidar for 3D mapping; radar; sonar; cameras; and GPS/GNSS for locating the vehicle. Other sensors such as gyroscopes and accelerometers can provide ‘dead reckoning’ navigation when there is no GNSS signal available, such as in tunnels. Those sensors are usually also used to detect if the vehicle is veering off course (eg, due to skidding on a slippery road), allowing the vehicle to take corrective action, and also to detect collisions (eg, to trigger airbags). The vehicle will probably also have sensors to detect the temperature, ambient light level (to control lights) and so on. It may even have a microphone to listen for the siren of an emergency vehicle, so it can pull over to let it pass. Lidar stands for Light Detection and Ranging. It is like radar, emitting laser pulses (rather than RF pulses, as in radar) from a rotating assembly to make a three-dimensional map (point cloud) of the environment based on the time for the reflected signal to return. An example of a commercial lidar device for ADAS or autonomous vehicles is the HESAI Automotive-Grade 120° Ultra-High Resolution LongRange lidar (siliconchip.au/link/ac6p) – see Figs.3 & 4. That model is said to acquire 34 million data points per second to a range of 300m. ADAS and autonomous vehicles usually have multiple cameras. The Figs.3 & 4: a HESAI lidar (Light Detection and Ranging) unit shown inset. Under it is an example of lidar imagery (a point cloud) with 128 channels (bottom) and the superior HESAI unit with 1440 channels (top). Source: www.hesaitech.com/product/at1440-360 14 Silicon Chip Australia's electronics magazine imagery from these has to be turned into meaningful data that can be used by the controller. This is done by software to create a three-dimensional map (point cloud), while extracting other useful data. An example of software used for this is Nodar Hammerhead (siliconchip.au/link/ac6s), shown in Fig.6. Sonar sensors use ultrasonic sound waves to measure distance, providing short-range information about objects in the immediate vicinity of the vehicle. Radar sensors use microwave radio beams to measure the range, velocity and direction of objects within their field of view. The sensors need to work regardless of conditions such as heavy snow, rain, ice, fog, road line markings being obscured or absent, changes in road surfaces, debris on the road, dirt roads etc. No single sensor is good at everything under all conditions, so a variety of sensors are needed. For Fig.5: environmental sensing by an autonomous vehicle with multiple cameras, radars, ultrasonic systems and a lidar unit. Original source: www.mdpi.com/1424-8220/23/6/3335 siliconchip.com.au example, the sensors shown in Fig.5 are fused to produce the capability shown in Fig.7. Software Hazard assessment Relevant software standards include ISO 26262, which is a process for managing and reducing risks for electrical and electronic systems in road vehicles. It covers planning, analysis, design, implementation, verification, validation, production, operation and decommissioning. It includes guidance on model-based development, software safety analysis, dependent failure analysis, fault tolerance and more. ASIL refers to Automotive Safety Integrity Level, a risk classification system specified by ISO 26262. It defines functional safety as “the absence of unreasonable risk due to hazards caused by malfunctioning behavior of electrical or electronic systems”. There are four levels of risk associated with system failure: A, B, C & D, with A being the lowest level and D the highest level of hazard if a system fails – see Fig.9. The higher the risk level, the greater the required reliability and robustness of the particular system. AEC-Q100 is a standard that ensures the safety of electronic parts used in cars, focusing on reliability stress-­ testing of integrated circuits. Fig.6: an actual image from the Nodar Hammerhead at upper left and the processed image outputs from their stereovision software at upper right and bottom. Source: www.nodarsensor.com/products/hammerhead Fig.7: the capabilities of the sensors from Fig.5 fused to show the overall detection capability for cameras, radar & lidar at lower right. Original source: www.mdpi.com/14248220/23/6/3335 Fig.8: the architecture of NVIDIA’s DriveOS software. ◀ According to Synopsys, today’s autonomous cars use 100 million lines of code, and in the near future, they will have 300 million lines. Operating systems for autonomous vehicles include QNX Neutrino (used by Acura, Audi, BMW and Ford among others; Unix-like); WindRiver VxWorks (also used by BMW, Ford and the Mars Perserverance rover); NVIDIA’s DriveOS (see Fig.8, used by Audi, Mercedes-Benz, Tesla and Veoneer); along with Integrity. Apple, Google and Microsoft also have their own versions of autonomous vehicle operating systems in use or under development. AUTOSAR AUTOSAR (AUTomotive Open System Architecture; www.autosar.org) is a global automotive and software industry partnership to develop and implement an open and standardised siliconchip.com.au Australia's electronics magazine October 2025  15 software, electrical and electronic framework for “intelligent mobility”. It defines things such as common interfaces, communications protocols, data formats etc. The layered architecture of AUTOSAR includes an application layer (vehicle specific), a runtime environment (that manages communications between software components), a basic software layer (communications and memory management, etc) and a control unit abstraction layer, to allow software to be developed regardless of specific hardware – see Fig.10. Fig.9: the ASIL hazard assessment levels for the failure of various systems on an autonomous vehicle. A indicates the least concern of failure, while D is of most concern. Source: www.synopsys.com/glossary/what-is-asil.html Advanced Driver Assistance Systems Fig.11: the Cruise self-driving car. Source: https://unsplash.com/photos/a-carthat-is-sitting-in-the-street-PkKsHQ5u4g8 These systems can help a human to operate a vehicle at SAE automation levels 0 through 5, or be integrated under the control of a master system to drive a vehicle autonomously. Unfortunately, the names of these features and their dashboard symbols are not always standardised between manufacturers. Adaptive Cruise Control is a system that automatically adjusts vehicle speed to maintain an appropriate separation from the vehicle in front. It uses sensors such as radar (typically at 24GHz or 77GHz), lidar or binocular cameras (eg, Subaru’s “EyeSight” system) to determine the distance to the car ahead. Adaptive Headlamps use a system to automatically adjust the headlight beam to avoid dazzling oncoming drivers (in theory, at least). The distance to oncoming drivers, if any, is estimated and the beam reach is adjusted appropriately. There is no binary high- or low-beam in some systems; just a continuously variable range. In one system by Mercedes, for example, the beam reach is adjusted between 65m and 300m, and adjustments are made every 40ms according to information from a vehicle camera that determines the distance to other vehicles. Anti-lock Braking Systems (ABSs) are designed to prevent a vehicle from skidding under hard braking, which can both result in longer stopping distances and make steering ineffective. It was originally introduced for rail vehicles in 1908 (although for a different purpose; to improve brake effectiveness), and 1920 for aircraft, but it was not universally adopted. The widespread adoption of ABS Australia's electronics magazine siliconchip.com.au Fig.10: the AUTOSAR software architecture, the acronyms stand for VFB: Virtual Functional Bus; RTE: Runtime Environment; BSW: Basic Software. Original source: Fürst, Simon “AUTOSAR – A Worldwide Standard is on the Road” – siliconchip.au/link/ac6t 16 Silicon Chip for aircraft happened in the 1950s. These were hydraulic systems, but an electronic system was developed for the Concorde in the 1960s. The modern ABS system for cars was invented in 1971 by Fiat and has been used on many models since then. It has been required on almost all cars sold for decades now. Modern systems monitor the rotational speed of each wheel and compare that with the speed of the vehicle. If one wheel is rotating slower than the rest of the vehicle, the brake pressure for that wheel is reduced, unless the car is turning. Brake pressure can be reduced or reapplied up to 15 times per second, and each wheel can be controlled individually. In more modern vehicles, the ABS system is also part of the electronic stability control system. Automatic Emergency Braking uses forward-looking vehicle sensors, such as radar and lidar, to sense the distance and time to impact of a vehicle or other obstacle. If the driver does not brake in time, the brakes are automatically applied. This might also be used in conjunction with automatic emergency steering (if fitted) if the braking distance is insufficient. Automatic Emergency Steering tries to steer a vehicle away from an imminent collision. Hazards that can be avoided include cars, cyclists, pedestrians, animals or road debris. Automatic emergency braking may also be implemented. Decisions are made based on inputs from radar, lidar, cameras, ultrasonic sensors etc. The process for action is: 1. Detection; continuous monitoring from sensors 2. Assessment; the control module uses data from the sensors to determine the vehicle velocity, trajectory, distance to the obstacle etc 3. Decision; if a collision is determined to be imminent and cannot be avoided by emergency braking alone, the calculations are made for a steering manoeuvre 4. Action; the steering actuator is activated by the control module to steer the vehicle on a path calculated to avoid the obstacle and any other obstacles 5. Notification; the driver is notified of the action There are various levels of Automated Parking, from basic to fully automatic. For automated parking to siliconchip.com.au ◀ Fig.12: the process of automatic parallel parking. Original source: “A novel control strategy of automatic parallel parking system based on Q-learning” – siliconchip.au/link/ac6u Fig.13 (below): possible parking scenarios for Volkswagen’s Parking Assist. Original source: Green Car Congress – siliconchip.au/link/ac6w work, the parking space needs to be ‘parameterised’ so that the appropriate vehicle direction, steering angle and speed can be computed – Fig.12 shows a reverse parking scenario. Other parking scenarios are possible, for example, right-angle parking. Volkswagen is one of many manufacturers who have developed automated parking, which they call “Parking Assist”, through three generations, plus fully automatic parking. Their first generation only allowed for reverse parking into parallel spaces, with a maximum of two moves, and the target space had to be 1.4m longer than the vehicle. Vacant parking spaces could be detected at up to 30km/h. It used ultrasonic sensors. Their second generation could perform multiple manoeuvres to park, as shown in Fig.13. It used cameras in the side mirrors, at the front and the rear, as well as ultrasonic sensors. The third generation could park the vehicle into a much smaller space and detect vacant spaces at speeds up to 40km/h. Australia's electronics magazine These Parking Assist modes correspond to SAE Level 1, and require driver supervision. Beyond that, Parking Assist at SAE Level 4 provides for fully automated parking with no human intervention required. Automated Valet Parking is a system developed by some manufacturers for a car to park and retrieve itself in certain parking garages. Infrastructure is required at the car park, as well as communication between the vehicle and the car park via V2X technology (see below) to receive instructions and location information within the car park. For more on this, see the video at https://youtu.be/30eB8Jj7xh0 Tesla also have an “Actually Smart Summon” feature, where the car will unpark itself and come to the driver with the use of a smartphone app as long as the car is within 65m of the driver, with a clear line of sight, and is not on a public road. Automatic Wipers: rain-sensing wipers were invented by an Australian Raymond J. Noack. Moisture is October 2025  17 windshield LED photodiode raindrop Fig.15: the output of the Tesla Driver Drowsiness Warning, which is not visible to the driver. Source: www.vehiclesuggest. com/tesla-hackerfigured-out-a-wayto-fool-tesla-camerabased-drivermonitoring-system Fig.14: the operation of an automotive rain sensor. In the presence of raindrops, there is some loss in the strength of the infrared beam reflected. Source: https://w.wiki/ERxC detected on the windscreen, and the wipers are activated at an appropriate speed and interval. The rain sensor is typically located in front of the rear-view mirror, and monitors infrared light reflected back from the outside surface of the glass, as per Fig.14. Blind Spot Monitors use radar or cameras to monitor a driver’s so-called ‘blind spot’ and provide a warning before they attempt to move into it if something is detected there (eg, a motorbike). Subaru’s EyeSight Camera system uses a pair of stereo cameras and was first launched in 1989. It is used for Adaptive Cruise Control, but can also provide sensory input for pre-collision braking that detects cars, motorcycles, bicycles and pedestrians. In the USA, the system was found to reduce rear-end crashes and injuries up to 85%. Subaru is working to integrate an AI judgement capability into its EyeSight system. Climate Control is a feature in most vehicles now, providing both heating and cooling. It is important for both safety and comfort, for example, to ensure that the windows remain clear while driving. Some cars have automatic defogging features, including some Kia and Hyundai models. Collision Avoidance System is a system that monitors a vehicle’s speed, the distance to the vehicle in front and its speed, to provide a warning or take corrective action if a collision is imminent. Sensors, such as radar and lidar, are used to determine vehicle parameters, like speed and distance. Automatic Emergency Braking and Automatic Emergency Steering are two possible systems that are used to implement collision avoidance. Crosswind Stability Control was first used by Mercedes Benz from 2009 in some cars, then later, vans and trucks. A deviation caused by crosswinds can be automatically corrected with the vehicle’s ESC system by several methods, such as steering, torque vectoring to provide more Fig.16: an algorithm flowchart for implementing electronic stability control (ESC). Original source: https:// autoelectricalsystems. wordpress.com/2015/12/20/ electronic-stabilityprogramme-esp 18 Silicon Chip Australia's electronics magazine drive force on the left or right side of the vehicle, or differential braking. Driver Drowsiness Detection uses cameras and sensors such as eye-­ tracking sensors to monitor driver behaviour and sound an alarm to alert the driver if drowsiness is detected. Drowsiness is detected by sensing behaviours such as yawning, eye blinking rate, eye gaze, head movements, facial expressions and driving behaviour, such as lane deviations and speed variations. Machine learning analyses behaviour patterns and learns to identify behaviours corresponding to drowsiness. The idea is to alert the driver to rest before they fall asleep. Tesla Driver Drowsiness Warning uses a camera to monitor the driver and sounds an alert if drowsiness is detected. Volkswagen monitors lane deviations and steering movements to detect drowsiness. Other companies offering this feature include BMW (Attention Assistant), Citroën (AFIL/LDWS), Jeep (Drowsy Driver Detection), Subaru (Driver Monitoring System), Toyota (Safety Sense) and Volvo (Driver Alert System). Others also include Ford, GM, Hyundai, Kia. Some fleet operators, such as trucking companies, install centrally monitored driver drowsiness detection systems in their vehicles, which are monitored using AI systems and/or humans. Fig.15 shows the output of a Tesla Driver Drowsiness Warning obtained by <at>greentheonly as he tests the camera with different scenarios such as “driver’s eyes nominal”, “driver’s eyes down/closed/up”, “view of head truncated”, “driver looking left/right”, “camera dark/blinded”, “driver head down”. You can see his video at https:// youtu.be/pZWR4MQBI4M siliconchip.com.au Driving Modes such as for snow, ice, sand, hill ascent and descent control etc are available on some vehicles. The vehicle’s performance is optimised via control algorithms with the throttle response, traction control, stability control, transmission behaviour etc, adjusted as required. Electronic Stability Control (ESC) expands on ABS by adding a steering angle sensor and a gyroscopic sensor. If the intended direction of the vehicle doesn’t correspond to the actual direction it is travelling (ie, it is losing traction), the ABS system can individually brake between one and three wheels to bring the vehicle back into alignment with its intended direction. The steering wheel sensor also provides information for Cornering Brake Control (CBC) to take into account the differential rotational speed of the wheels on the inside and outside of the curve. A typical control algorithm is shown in Fig.16. Heads-up displays (HUDs) convey information to the driver, such as speed, the current speed limit, the distance to the vehicle ahead, turns for navigation etc. This information is projected onto the windscreen; see Fig.17. Ice Warning is important in colder climates as ice is often not visible on the road (‘black ice’) and this is a serious safety hazard. A variety of detection systems are used, such as multispectral imaging systems, to examine the road surface; thermal imaging systems; air temperature and humidity measurement; weather data from external sources; or information from vehicleto-­infrastructure (V2I) or vehicle-to-­ vehicle (V2V) systems. Intelligent Speed Adaption (ISA) is a system that reads road signs or uses other data to ensure that the driver stays within the speed limit for that section of road. There may be a warning if the driver exceeds the limit, or the driver may be able to request the car travels at or below the limit. Intersection Assistance is when a vehicle is equipped with side-looking radar to detect if drivers are coming at right angles to the car; brakes can be automatically activated to avoid a collision. Lane Deviation (or Departure) Warning uses cameras to monitor lane markings, to warn a driver if they start to depart from the lane they are in, or to siliconchip.com.au Fig.17: a head-up display rendering showing various ADAS parameters. Source: www.eetimes.com/add-ar-displays-for-adas-safety Fig.18: the Night Vision Assistant on an Audi A8. Source: https://w.wiki/ERxE Fig.19: the live 360° camera view on a Mazda CX-9. keep them in the centre of the lane even if they are not actively steering the vehicle. Lane Change Assistance uses sensors to detect if vehicles are in the driver’s blind spots, and will alert the driver if they are. Navigation in an ADAS vehicle may involve route recommendations or alternatives, choice of toll or no toll roads, advice on traffic congestion Australia's electronics magazine etc. The vehicle may receive real-time updates as conditions change, such as traffic congestion forming. Position information is obtained with GPS or another GNSS system. Night Vision is a system using infrared cameras to improve driver awareness at night or in poor conditions – see Fig.18. The first car to be offered with this technology was the 2000 Cadillac de Ville. October 2025  19 Do autonomous cars get confused? This short video shows Waymo cars honking at each other: https://youtube. com/shorts/PkVSoTZBh8U This video shows a Waymo vehicle not taking the passenger where they wanted to go on a simple trip: https://youtu.be/-Rxvl3INKSg A police officer pulls over a Waymo: https://youtu.be/7W-VneUv8Gk Omniview is a type of camera system that gives a 360° and/or bird’seye view of a vehicle. It is known by many other names, such as Surround View. It was first introduced as on the 2007 Nissan Elgrand and Infinit EX, as “Around View Monitor”. Video feeds from four to eight cameras are synthesised into a bird’s-eye view to assist drivers with park, or to remotely view their vehicle and its surrounds – see Fig.19. There is quite a bit of processing required to convert the images from the cameras into a (mostly) seamless 360° image. The steps include: 1. resizing the images 2. removing lens distortion 3. perspective transformation 4. stitching the images together 5. displaying the results Such systems can also be retrofitted. One example we found is the Taffio 360° Surround View Set (siliconchip. au/link/ac6q). Parking Sensors are usually ultrasonic rangefinders that give the driver an audible (and visual) indication of how close they are to objects. Typically, the closer the vehicle is to an object, the faster it beeps. These systems are often accompanied by a rear-facing camera, which may have lines marked on the image to assist the driver with determining the path of the vehicle in relation to obstacles. Reversing Cameras are a common feature now (required in new cars) and a relatively simple one to implement. The first known vehicle reversing camera was on the 1956 Buick Centurion concept car. The first commercially produced car to have one was the 1987 Toyota Crown in the Japanese market. Temperature Sensors are used to measure inside and outside temperatures, and may contribute to ice warning data or the operation of the climate control system. Traction Control is a system to ensure that wheels don’t lose traction with the road during heavy acceleration. Each wheel has a speed sensor, and the speed data is sent to the ECU, which compares it with the speed of 20 Silicon Chip the vehicle. If there is a mismatch, taking into account if the car is cornering or not, the engine torque is reduced or a brake is applied on the wheel. Traffic Jam Assist is a feature that uses Adaptive Cruise Control and Lane Departure Warning to take over driving in traffic jams. A safe distance is maintained with the vehicle in front. Traffic Sign Recognition uses a camera to recognise traffic signs, such as stop and speed limit signs, giving appropriate warnings to drivers. Traffic sign recognition is facilitated by the Vienna Convention on Road Signs and Signals, which has attempted to standardise road signs across various countries, although Australia is not a signatory. Traffic sign recognition systems use a variety of different algorithms, such as recognising the board shape and using character recognition to read the writing. A further level of complexity uses convolutional neural networks (CNN), which are trained with real signage and use deep learning to recognise various signs. The output of the Freeman Chain Code and shape determination of the algorithm can also be used as an input to CNNs. A typical sign recognition algorithm includes the following steps: 1. capture an image of the sign(s) with a colour camera 2. convert the image from RGB to HSL (hue, saturation, lightness) 3. apply a Gaussian smoothing filter 4. detect edges using a Canny edge detector algorithm 5. use a Freeman chain code algorithm to detect letters and numbers 6. use a polygonal approximation of digital curves to detect the sign shape 7. display the result Tyre Pressure Monitors use either inferences from other data or direct pressure measurements. For indirect systems, parameters such as wheel speeds, accelerometer outputs and other vehicle data are used to make inferences about tyre pressure, and a warning is issued to the driver to check pressures. Australia's electronics magazine That is not as accurate as direct measurement systems, which use a sensor in each wheel to determine the pressure. The sensor may either be battery-­ operated, which requires maintenance to replace the battery, or may be wirelessly supplied with power like RFID systems. Wrong Way Driving Warning is a system on some vehicles to alert the driver if they are driving in a direction which they are not meant to, as determined by GPS data. It doesn’t seem to be widely implemented. V2X stands for vehicle-to-everything and describes wireless communication between the vehicle and any other vehicle or entity with which the vehicle may interact. Vehicle to infrastructure (V2I) and vehicle to vehicle (V2V) are related systems. Operational Design Domain The Operational Design Domain (ODD) defines the set of conditions such as environmental, geographic, time of day etc under which the vehicle is certified to operate safely. In other words, it is a recognition of the limitations of the autonomous system. If the situation in which the vehicle finds itself is outside of the ODD; for example, certain traffic or road conditions, it might warn the driver or passenger and deactivate itself to allow the driver to assume control. Alternatively, the vehicle may park itself. Various standards and regulators have defined the exact meaning of ODD. An example is Mercedes Benz stating the following for its Drive Pilot Level 3 system for supervised autonomous driving, which is certified for use in California and Nevada: ...requires speeds below 40 miles per hour, clear lane markings, not too much road curvature, clear weather and lighting conditions, and a high-definition map to be available in the system’s memory... Warning sounds Electric autonomous vehicles can be so quiet that pedestrians may not hear them, so they are required to make a sound at lower speeds. In Australia, as of November 2025, all new electric, hybrid and hydrogen-powered cars, buses and trucks will be required to be fitted with noise-making systems which make a noise of 50dB below 20km/h. Similar laws apply in the EU, Japan, the UK and the USA. siliconchip.com.au Legal liability for accidents For SAE levels 0-3, the driver must be able to take control of the vehicle at any time, and they will be liable for any accidents, as they should be constantly monitoring the vehicle, ready to take control at any time. For levels 4 & 5 vehicles, there is no “driver”; they might not even have any access to vehicle controls. It is unclear who would be responsible for an accident that may occur. Fully autonomous vehicles We will now look at examples of autonomous vehicles, starting with one from Australia. Australian road trains Australian company Mineral Resources (www.mineralresources. com.au; MinRes) developed worldfirst autonomous road trains that can haul 330 tonnes of iron ore along 150km of private road in Western Australia, from the Ken’s Bore mine site to the Port of Ashburton. The trucks are converted Kenworth models. There are 150 trucks in the fleet, and they drive at 80km/h. There is an interval of 2-3 minutes between each truck as they constantly run along the road delivering iron ore. Hexagon (https://hexagon.com) performed the conversions – see Fig.20. According to their description, this includes: a sensory system for awareness (truck performance, surroundings and location); an autonomy layer, the brains for decision making; and a by-wire system for controlling the vehicle. Table 1 – Tesla autopilot features (source: https://w.wiki/3wkp) Feature Autopilot Enhanced Autopilot Full Self Driving Traffic-aware cruise control Autosteer Navigate on autopilot Auto lane change Autopark Summon Smart summon Traffic & stop sign control Autosteer on city streets ✔ ✔ ✖ ✖ ✖ ✖ ✖ ✖ ✖ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✖ ✖ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ Fig.20: the world’s first autonomous road train, in Australia. Source: www. mineralresources.com.au/our-business/onslow-iron-project/autonomous-roadtrains Buses and shuttles The Apalong is a Level 4 driverless bus from China that has been in production since 2017 – see Fig.21. It travels at between 20km/h and 40km/h and can accommodate 14 people. It uses Baidu’s Apollo 3 Open Driving Platform (https://github.com/ ApolloAuto/­apollo). Cars Tesla is constantly updating the software in its vehicles. It has a feature called “Autopilot” or “Enhanced Autopilot” available in all its cars produced since 2019, as well as some vehicles offering “Full Self-Driving” (FSD; supervised). The capabilities of different versions of the software depends on the siliconchip.com.au Fig.21: the Apalong autonomous bus from China. Source: https://w.wiki/ERxF Autonomous vehicle software Few manufacturers have released the code for the autonomous cars, but the Stanford Racing Team, the progenitor of Waymo One, released the code for the vehicle that won the 2005 DARPA Grand Challenge event at: https://sourceforge.net/projects/stanforddriving/ The vehicle ran this code, written in C and C++/ on a Linux operating system running on Pentium M CPUs. Australia's electronics magazine October 2025  21 Fig.22: a Tesla Hardware 3 (HW3) Full Self Driving (FSD) board. A lot of the circuitry at the top and bottom is the power supply for the two large UBQ01B0 multi-core processors. Source: https://w.wiki/ERxG Fig.23: an autonomous mining truck for transporting minerals. Source: Fortescue Metals Group Ltd – www.mining-technology.com/features/australialeads-the-way-in-autonomous-truck-use market and local laws. Tesla classifies these systems as SAE Level 2, possibly for legal reasons, as FSD is arguably a Level 4 technology (see Fig.26). The FSD v12 software is available for later vehicles with Hardware 4 (HW4; in Model S and Model Y after January 2023). It uses a neural network and artificial intelligence that has been trained on millions of video clips. Older versions of the code were reliant upon rule-based algorithms written in C++, but later versions now use an ‘end-toend’ neural network that constantly learns and adapts. End-to-end means that the entire FSD system is a neural network, not just parts of it. The high-level Python programming language is used for machine learning, with C++ for embedded systems. The software all runs under the Linux operating system. Samsung makes the processor for HW4, a custom ‘system on a chip’ (SoC) device that has 16GB of RAM and 256GB of storage. The internals of the HW4 computer can be seen at: siliconchip.au/link/ac6l siliconchip.au/link/ac6m The second link states that HW4 is running Linux kernel 5.4.161-rt67 Fig.22 shows a Tesla FSD board. We can see that the main chips are labelled UBQ0180. Wikichip (see siliconchip. au/link/ac6n) states these are FSD chips that incorporate “3 quad-core Cortex-A72 clusters for a total of 12 CPUs operating at 2.2 GHz, a Mali G71 MP12 GPU operating 1 GHz, 2 neural processing units operating at 2 GHz, and various other hardware accelerators. The FSD supports up to 128-bit LPDDR4-4266 memory”. Each chip contains 6 billion transistors. As it was first shipped in Teslas in 2019, we believe this unidentified board is a Hardware 3 or HW3 board. Table 1 illustrates the capabilities of Tesla’s Autopilot, Enhanced Autopilot and Full Self Driving. Fig.24: the Liebherr T 264 battery-electric autonomous mining truck, jointly developed with Fortescue. Source: Liebherr – siliconchip.au/link/ac6v Mining vehicles Australia is the world leader in the use of autonomous mining trucks – see Fig.23. As of May 2021, we had 575 such vehicles, compared to 143 in Canada, 18 in Chile, 14 in Brazil, 12 in China, 7 in Russia, 6 in Norway, 5 in the USA and 3 in Ukraine. Fortescue and Liebherr jointly developed an autonomous battery-­electric Australia's electronics magazine siliconchip.com.au 22 Silicon Chip T 264 truck, resulting in an order for 475 Liebherr machines. The T 264 is 8.6m wide, 14.2m long, 7.2m high with the dump body on and can carry a payload of 240 tonnes. The truck itself weighs 176 tonnes. The prototype truck (Fig.24) has a 1.4MWh battery weighing 15 tonnes that’s 3.6m long, 1.6m wide and 2.4m high. It’s made up of eight sub-packs, each consisting of 36 modules. It can regeneratively charge as it goes downhill. Taxis Waymo One (https://waymo.com/ waymo-one) is an autonomous taxi service currently available in the US cities of Austin, Los Angeles, Phoenix, San Francisco and soon Atlanta and Miami. Waymo One is a subsidiary of Alphabet Inc, Google’s parent company. Waymo vehicles have been under development since 2015, and in 2020 offered the self-driving service without safety drivers present in the car. The company traces its origins to the 2005 and 2007 US Defense Advanced Project Agency’s (DARPA) Grand Challenge competitions and the Stanford Racing Team. They won first place in 2005 and second in 2007. Waymo have applied their self-­ driving technology to several vehicle platforms; currently they use Jaguar I-Pace EVs (Fig.25), with whom they have a partnership, at an estimated additional cost of US$100,000 ($156,000) per vehicle. As of May 2025, approximately 1500 autonomous Waymo One vehicles were in service, mostly the I-Pace. Waymo vehicles are twice as safe as human drivers according to accident statistics, but have nevertheless been involved in incidents, mostly minor. A Waymo One taxi can be summoned via an App. Amazon’s Zoox (https://zoox.com) could be considered a ‘competitor’ to Waymo. They are also an autonomous taxi service operating in California and Las Vegas, Nevada. Their vehicles are fully electric and have no steering wheel (see Fig.27). Fig.25: Waymo’s modified Jaguar I-Pace EV. I-Paces have been discontinued, but Waymo acquired a large number and continues to deploy them. Source: https:// waymo.com/blog/2018/03/meet-our-newest-self-driving-vehicle Fig.26: a screenshot taken from an example video of Tesla’s FSD (Full-Self Driving). Source: www.tesla.com/fsd Further reading More details on some of these ADAS systems can be seen in our features on Automotive Electronics, December 2020 and January 2021 (siliconchip. SC au/Series/353). siliconchip.com.au Fig.27: an Amazon Zoox robotaxi, which is design as a fully-autonomous taxi (see https://zoox.com). Source: https://w.wiki/ESSv Australia's electronics magazine October 2025  23