Three years ago, we all believed that autonomous cars were just around the corner – and we felt that was a good thing. Retain human control for the fabled blast in the country, but let the machines take on the drudgery of commuting and long-distance motorway work – and getting us home from the pub!
So what’s changed?
It was shocking to read, in Autocar, that Andy Palmer, boss of recently floated Aston Martin, quoted as saying “The idea of full autonomy being widespread in my lifetime is absurd. Full Level 5 systems are a moonshot.” As an aside, Mr Palmer was also scathing about Brexit, confirming that the delay was the worst of all worlds, preferring a decision, any decision, to be made to close down the uncertainty.
We have some doubts about the direction of Aston Martin Lagonda Ltd as a company, but we respect Andy Palmer as a well-connected, senior car-industry figure. So when he says that full autonomy is a pipe-dream, we listen.
Another issue revealed to us is the autonomy “big accident” risk. When autonomous cars crash, they tend to be large ones! If humans have crashes, mostly they will realise that something has gone wrong and slam on the brakes at the last split-second. Whilst this does not prevent the crash, it does mean that some deceleration occurs before impact, and so the crash happens at much reduced speed. Typically, crashes in autonomous cars happen because of a failure to correctly interpret the surroundings of the vehicle. And thus the autonomous car hasn’t noticed anything wrong – so it ploughs into the obstruction at full speed! Ooops.
There have been two famous crashes, where Teslas have sped straight into the side of juggernauts parked across the highway. Speculation among the online community (oh dear, not the most reliable source then) is that the crashes happened because seeing a juggernaut sideways is such an unusual occurrence that the AI-developed software interprets the sight as an overpass bridge and so ignores it.
Finally, there is the fabled issue of how can one let a computer decide whether to swerve away from a man in the road, if that then endangers two children on the pavement.
However, we are not convinced that these issues will prevent autonomous cars. At its present stage of development, it appears that sensors, processing power and software are not quite there. But these are engineering problems that are easy to define and will be solved.
Computers are very good at measuring distances and heights. Therefore, interpreting a truck as a truck and not a bridge can be pre-programmed. Likewise, new types of sensors will be developed so that cars will know much better than human drivers what is going on around the vehicle. Add to that car-to-car connectivity, so that each car knows the intention of all the vehicles within half a mile, and suddenly an autonomous car is much better placed to co-ordinate its movements with those of all the surrounding cars.
We do not agree with the idea that humans are better at split-second, morally loaded decisions than computers. In a crash situation, the choices made by a driver will be essentially random, or pre-programmed by their normal reaction. Given a little forethought by the software programmers, 99.9% of situations can be managed for an optimal result rather than the vague human output.
Convoys of communicating vehicles can travel closely together, allowing for more efficient use of roads, and greater fuel efficiency.
- We wonder if Mr Palmer’s reluctance over self-driving cars is that it removes a key justification for buying an Aston Martin – and that providing such systems is also beyond the capacity of a relatively tiny car company?
- However, combining human drivers with convoys of autonomous vehicles could be tricky.
It is this last point which is exercising us! We believe that fully autonomous roads will happen within the next 10 years. Our worry is that will the take over of roads by autonomous vehicles mean that car enthusiasts in their old-fashioned, petrol engined “classics” are banned from going out at all eventually?