Autonomous Driving – Slough of Despond.

Life in autonomy remains dreadful.

• There is little hope for a recovery in autonomous driving anytime soon as programs are being cancelled meaning that my 2028 target looks increasingly over optimistic with every day that passes.
• All that anyone is willing to commit to (or talk about at trade shows) these days is eyes-on driving down the highway and those that are still going for full autonomy will no longer commit to a time frame.
• The latest to fall appears to be GM which may be cancelling its Ultra Cruise product probably triggered by the disaster that GM has been left to deal with after the road accident involving a Cruise vehicle (see here).
• In this article I concluded that the issue with the regulator was a misunderstanding, but it appears that I was not correct.
• The fact that Cruise / GM has fired a number of employees who were involved in communicating with the regulator implies that Cruise wilfully misrepresented the incident which is a much more serious turn of events.
• Consequently, the reputation of both Cruise and GM in autonomous driving has been severely tarnished which is what may have led to a decision to cancel Ultra Cruise.
• Ultra Cruise is the next step on from Super Cruise which combines adaptive cruise control, lane centring and forward collision mitigation to allow a vehicle to drive by itself down the highway.
• Currently, Super Cruise only works on highways where the vehicle has a high-definition map on board and where the driver is paying attention.
• Ultra Cruise is supposed to build on this with a richer sensor suite and the ability for the vehicle to drive itself 95% of the time (highways and other roads) as long as driver is paying attention.
• This is supposed to be another stepping stone towards full autonomous driving but if the product is cancelled, it will set GM back even further.
• GM’s problems are not occurring in isolation and the problem that I think that even the leaders (Waymo, Mobileye, Baidu) have is that they have not addressed the root of the problem.
• This problem is caused by basing machine vision purely on deep learning systems which are incapable of dealing with anything that they have not seen in the training data.
• This is fine when a task is well-defined and nothing changes but the road has neither of these characteristics which is why autonomous driving systems continue to fail to drive better than humans.
• I continue to think that it is not until this issue is properly addressed that autonomous driving has a chance of hitting real commercialisation.
• There has not been much progress on this front although I have long suspected that the answer lies in combining both deep learning and ordinary software as their strengths and weaknesses are highly complimentary.
• In the meantime, it is increasingly looking like it will be a long wait meaning that I am going to be forced to change my 2028 target.
• Valuations also remain at rock bottom meaning that anyone who is brave and can judge who will eventually win is in line to receive a bargain.
• The closest I have got to this sector is Ouster which does not rely on autonomous driving to underpin its growth but rather the growth in industrial robotics and automation.