Tesla Autopilot updated

Earlier this year, on the 7th of May to be precise, a Tesla Model S crashed into a truck trailer killing its driver. The catch was that the driver wasn’t driving, Tesla’s autopilot system had complete control of the vehicle. While Tesla have made it abundantly clear that the Autopilot system is not ready to be given complete autonomous control of a car but this has not stopped an enquiry into the company.

A diagram of the Tesla Crash taken from the New York Times.

A diagram of the Tesla Crash taken from the New York Times.

Since the incident Tesla have unveiled some changes that both deal with and explain why the Autopilot system did not recognise the trailer. Firstly, the way the world looks to a radar sensor plays a big role in the difficulties Tesla is facing. People, wood, animals, in general any organic matter is almost transparent to radar as the radio waves emitted go right through these softer materials, instead of being reflected back to the vehicle’s sensors. Another issue is that metallic, concave objects not only reflect the radio waves, but also magnify them making these objects look much bigger:

” A discarded soda can on the road, with its concave bottom facing towards you can appear to be a large and dangerous obstacle, but you would definitely not want to slam on the brakes to avoid it.”

The issue is how to identify dangerous obstacles while avoiding slamming on the brakes unnecessarily. The biggest change being made to adapt to this problem is fleet learning. Every Tesla that drives along a given route will be feeding data about it to other Teslas, making routes that are the most traveled the safest. Coupling this with upgraded radar software, capable of mapping the surrounding area from multiple snapshots (the previous system relied on a camera feed and a single snapshot) will easily spot obstacles that are moving against those that are static. There is still room for error, of course, with Teslas relying on data that was collected by other vehicles in an inconstant environment.

For the time being the only solution has come in the form of gradual breaking: if the system is uncertain about an obstacle ahead it will apply the breaks only slightly, reducing speed to gather more information, ideally it will come to a halt before the object but in the case it does not the vehicle will not have been travelling very quickly.

Autonomous driving has some way to go yet. Google and Tesla, among many others, are working full-time to get this technology ready for widespread use, we could be seeing fully autonomous personal vehicles before the end of the decade!

Quotes taken from Tesla