By Sandeep Sovani, director, Global Automotive Industry, ANSYS

The autonomous car is a promise that has been present for some years, but still isn’t a reality. Before autonomous cars are allowed on the roads, manufacturers must consider any problem that may arise and implement safety measures around them.

To guarantee safe vehicle, it is estimated that at least nine billion kilometres must be driven. In practice, this is infeasible. For example, the seven Google self-driving cars have driven only a few million kilometres in nine years.

This would mean that hundreds of years would be needed to get to that nine billion of test kilometres. The solution for this is simulation.

 

Importance of sensors

Autonomous vehicles use sensors and radars as their ‘eyes’ on the road. Each car needs around 20 or more to be effective. But if humans drive using two eyes (or sensors) – why does the car need so many?

The main reason is that the brain is more sophisticated and can understand its surroundings better than sensors can. However, in many ways, autonomous vehicles can be better drivers than people, as they can ‘telepathically’ speak to each other to understand the road and its surroundings.

However, to be better than humans, they need the right ‘training’ and to be considered trustworthy in unpredictable moments.

There are many situations that we cannot predict and that may or may not happen, such as a child suddenly crossing the road to run after a ball, but what manufacturers must guarantee is that the car is able to respond correctly to this type of situation.

Of course, there are billions of such scenarios and we can’t test them all, but with simulation, the sensors can be taught tor recognise different conditions so that the car makes the right choice.

 

Training sensors and choices

One of the challenges is getting the sensors on the vehicle to behave as you expect them to.

A human driver makes a million unconscious adjustments and adaptations to road conditions, which computers are unable to do, unless programmed to – and autonomous vehicles must do the same.

For example, one challenge is harsh weather conditions such as heavy rain. Rain is difficult to quantify – a computer can activate ‘wet weather mode’ to compensate for environmental conditions, but what counts as ‘heavy rain’ will vary.

Rather than carrying out extensive road-testing, companies can use simulation to train and evolve vehicle sensor technology, continually improving its response to ambiguous situations before carrying out real-world testing.

In a simulation, cars can fail and learn from it without anyone being hurt, and that is far safer than an accident happening in the real world.

Simulation is the key

Getting this correct is a challenge — a human challenge, and until we implement measures that ensure the autonomous car is safe, we can’t have full faith in them.

The industry needs to get smart and start using simulation to push forward the success of autonomous cars.