murraypaul wrote:ReformedCharacter wrote:I was just thinking to myself how hard a problem autonomous driving is.
I think about that on my drive to work.
T-junctions, roundabouts with and without lights, a level crossing, roads that start off with two normal lanes then narrow until two cars could pass, but not a car and a truck. Central lines that appear and disappear. Edge markings that appear and disappear. Roads with crumbling edges. Zebra crossings. Temporary lane closures, roadworks, parked cars.
I just don't think we are even close to a smart car that could navigate these things.
IMO, the future of autonomous driving will be smart roads, not just smart cars.
Our roads and road signage are designed for human drivers. Robot cars need roads designed for robot cars, that can directly communicate the information they need, and mediate between them.
What has been achieved so far is pretty decent, with autonomous vehicles trawling complex city streets on a daily basis virtually without incident. Obviously it makes sense to optimise roads for future autonomy, although in many cases that would result in road markings actually becoming less important, except maybe as a backup. They are there for vision dependent systems, but computers are telepathic. We are probably in the equivalent of the dodgy L2+ zone where the tech is being forced to cooperate with an existing hierarchy which is ill equipped to perform the tasks being left to it. If autonomous can handle anthropocentric road design, then the task will only get easier as road design is modernised and the focus shifts to enhancing the performance of AD's. Ultimately road design will be unencumbered by the limitations of a human driver, and the entire road design will be better optimised for autonomy.
Smart infrastructure will obviously multiply the benefits of AD. An early example was the traffic lights in Vegas which communicated their state and time to next change to Delphi's autonomous cars, something which is now becoming more widespread with compliant units already available in cars on the market today and some pedestrian lights containing a countdown timer. Another obvious opportunity would be motorway gantry speed displays. This is one of the reasons an ideological adherence to vision only is so misguided, it throws away many of the potential improvements which technology can bring to road transport.
This is a two way street with autonomous driving also making smart road management more effective, for example having cars which actually obey dynamic speed limits designed to maintain a constant optimal flow rather than racing each other between traffic jams Throw in car to car communication and you can have pre-negotiated lane merging without reducing both lanes to a crawl, A benefit already available from this technology are systems like those in the latest Golf which will warn you if a suitably equipped Emergency vehicle is approaching, or if there is a similarly equipped stationary car obstructing the road around the next bend.
Now throw in AI oversight to balance traffic volumes across multiple routes...
Parked cars are already, for most developers a solved problem. If Tesla and American highway emergency response units both had 5g beacon compatibility then even they might stop meeting in such awkward circumstances. Every day autonomous cars successfully pass tens of thousands of parked cars. Years ago I watched a lecture by (IIRC) Karl Iagnemma in which he talked about detecting whether or not the rear gate of a parked lorry was open or closed in order to inform the likelihood that it would pull off or that there would be people stepping in to the road. It's worth remembering that the most seen attempt at autonomous is far from the most sophisticated or the most successful.