No, Autonomous Vehicles Makers Have Not Figured It Out
I am reading article after article now where the press assumes that because these companies now admit developing a driverless vehicle is harder than they thought, will takes decades to finish, are developing in much more benign and safer geofences and they are using much more simulation that they have figured out where they went wrong. Sorry that is not the case. All they have done is rearrange the deck chairs. The original issues remain. Public shadow/safety driving can never lead to close to L4 and takes lives needlessly. At some point they will have to approach more complex scenarios and eventually experience thousands of accident scenarios thousands of times over. What do you think that will lead to? (Tesla is showing this now. Their geofence is all highway. Much, much harder than what other AV makers are running in now.) The paradigm shift that is needed has not occurred. That paradigm shift being in two parts. Not parts of two parts — TWO WHOLE PARTS.
The first being that 99.9% of public shadow and safety driving needs to switch to simulation. The second part being that simulation cannot be gaming based or pretty much anything being produced in the industry right now. What is needed is aerospace/DoD simulation technology. Why? Because the current systems are misleading these developers in to a false sense of security. As soon as any part of the system, any model (vehicles, tires, roads, sensors, the environment etc) pushes any of the performance curves of what they are currently using the ML will create a plan based on flawed information. A situation they will likely not find until an analogous real-world tragedy hits. When that comes the rework and litigation will put them out of business. Proper real-time and very precise models are required. If you want to test my POV ask your vendors or sim staff to show you the performance curves of that vehicle, the tires, the roads and sensors under various conditions. Now ask them to show you the full motion simulator you need to replace that real vehicle, in core scenarios, and that the latency added when it is in operation, under full load is 16msec or less.
Please see more in my articles here
SAE Autonomous Vehicle Engineering Magazine — End Public Shadow/Safety Driving
Common Misconceptions about Aerospace/DoD/FAA Simulation for Autonomous Vehicles
The Hype of Geofencing for Autonomous Vehicles
