HOW AUTONOMOUS CARS SEE THE WORLD
BASIC DRIVER ASSISTANCE
Modern cruise control rst appeared in a production car in 1958 and is considered the initial step towards some form of driving assistance. By the mid-1990s, adaptive cruise control (ACC) systems appeared, using radar or LIDAR to detect a slower vehicle ahead, reduce speed and, if required, warn the driver to hit the brakes. Following the introduction of ESP stability control – which included an automated braking function – ACC was upgraded to include automated braking.with ongoing development and re nement, this is still the basis of the ACC tted to many current vehicles. Around the same time, camera or sensor-based lane departure warning systems (LDWS) were introduced, with variations that included the ability to steer or brake individual wheels to keep the vehicle within the driving lane.these rely on lane markings to detect vehicle position, and early systems quickly disengaged when road markings were inadequate. However, the development of advanced lane-detection algorithms has improved their capability when lane markings are indistinct or otherwise compromised.
Used separately or together, these systems are highly rated by drivers for minimising fatigue and have helped to reduce accidents, and are now standard tment in many mid- to higher-priced vehicles.
01 Instrument display and illuminated buttons on the steering wheel rim indicate the Level 3 Mercedes-benz Drive Pilot is engaged.
DRIVER MUST MAINTAIN CONTROL
These arrangements, while providing increased driver convenience and safety, do require the driver to always maintain control of the vehicle. LDWS requires all drivers to constantly keep a hand on the wheel and monitors this by measuring the input torque that is created. Driver-facing cameras are also increasingly being used to monitor driver eye direction, and both systems warn the driver to take proper control if inattention is detected.
For example, in ideal conditions, with light traf c and vehicles cruising at reasonably similar speeds on a well-marked road, the vehicle can travel considerable distances without driver intervention. However, if an incident ahead required a lane change to avoid it, neither system would be capable of doing so and the driver’s intervention would be required.
It is this requirement for the driver to always be in control that distinguishes these support features from the automated features required for fully autonomous driving.
02 Mercedes-benz Drive Pilot is certified as a Level 3 system for use on selected highways in Germany and is expected to be rolled out to additional markets later this year.
LEVELS OF DRIVING AUTOMATION
The SAE and ISO have collaborated to provide standardised de nitions of the different levels of driving automation (see attached table). These range from Level 0 for features limited to providing warnings or momentary assistance, to Level 6,0 where the vehicle is required to be fully capable of driving itself everywhere and in all conditions: complete autonomy. The key point in this table is the split between Level 2 and Level 3, with the driver deemed to be driving up to Level 2 and not driving from Level 3, even if present in the driver’s seat. This step is proving to be the major hurdle, as no doubt will be the step to Level 4, where there can be no fallback to a driver.
In practical terms, Level 2 can be simply understood as driving on a highway, using ACC and lane-keeping assist (LKAS) at the same time to respectively control braking and acceleration and follow the vehicle ahead, while controlling steering to maintain position in the centre of the selected lane. The driver is still in control, is required to have their eyes on the road, and must be ready to intervene should anything beyond the scope of these two systems occur.
At Level 3, the vehicle is required to assume full control within prescribed operating conditions, requesting the driver to intervene only when it judges this is necessary. In theory, the driver is free to take their eyes off the road and do something else but must remain ready to assume control if needed.
At the time of writing, just two vehicle manufacturers, Honda and Mercedes-benz, have introduced Level 3 systems. Both of these systems are intended for use in heavy traf c freeway conditions only, at speeds below 60 km/h, and limited to a network of pre-mapped roads, stored onboard and updated regularly. Some manufacturers are close to following suit, while others consider it preferable to move directly from Level 2 to Level 4. Their concern is that following a period of uninterrupted automated driving, drivers may become complacent and develop “automation bias”, meaning they may not be ready to take control if required to do so.
RELIANCE ON DATA
Autonomous vehicles rely on a combination of sensors – primarily LIDAR, radar, ultrasonic and cameras – to be the “eyes” of the car and map surroundings, measure distances and speeds, and identify objects like other cars, traf c lights, pedestrians and road markings.
Up to Level 2, the vehicle is required to make straightforward responses to inputs received from its onboard sensors: braking, accelerating, turning left, or turning right. From Level 3 onwards, there is an exponential increase in the amount of data required, the complexity of interpreting it, and then deciding how the vehicle should respond. The detailed mapping of roads provides much of the basic information regarding lanes, gradients, junctions, speed limits, danger zones and so on, but the challenge lies with the variable factors, which need to be captured in real time from the vehicle’s sensing systems. these include temporary roadworks and other obstructions, along with deviations, pointsmen, emergency vehicles and the actions of pedestrians.
Anticipating the actions of pedestrians is a good example. The vehicle’s onboard sensors monitor nearby pedestrians, noting whether they are already stationary at a crossing or approaching it, whether they are looking towards oncoming traf c or at the crossing signals, or whether they are looking down at a cellphone and thus distracted.
Using all of this information, while simultaneously monitoring other traf c and traf c lights, the vehicle would now decide whether to proceed normally, slow down and be prepared to stop, or make an emergency stop … exactly the same observations and judgments any good driver would make in the same situation.
As could be imagined, the number of different permutations and scenarios is almost in nite, and manufacturers have reported carried out millions of kilometres of testing to gather data used in conjunction with algorithms and arti cial intelligence to make driving decisions. Tesla also makes use of a “shadow” function on its vehicles, comparing drivers’ responses in normal driving to their system’s planned responses, to further re ne their capabilities.
READY FOR PRODUCTION?
Despite this level of testing, there are still reports of autonomous vehicles sometimes behaving erratically and being involved in accidents, some of which have resulted in fatalities. While a few have been the result of system shortcomings, others can be attributed to drivers relying on Level 2 vehicles to drive themselves while not being in the driver’s seat and/or ready to intervene when required. Some recent studies have indicated that in the US, the crash rate per kilometre for autonomous vehicles is lower than the national rate, and injuries were also less severe. While the sample size of autonomous vehicles was too low for the study to be de nitive, it does indicate the signi cant progress already made.
Despite the advances in technology and the clear bene ts, a majority of US adults have indicated they would not yet feel comfortable driving in a fully autonomous vehicle. The concerns most frequently raised surround liability in the event of an accident – vehicle manufacturer or owner – and the risk of computer hacking taking over vehicle control. Based on the technological progress to date, most industry experts are con dently predicting safe autonomous vehicles will be available within the decade; they’ll just have to convince the public.