No. of Recommendations: 11
It’s a waste of time and money, unless you embed a tag in every human and dog.
Through my day job I spend a fair amount of time working with others that have brought , and continue to bring, semi-autonomous machines to market. What I think everyone is describing is the truly amazing ability the human being has to sense and react to its surroundings in a highly effective manner. Beacons, consistent lines, buried cables, detailed maps, satellite positioning... are all tools that make an "autonomous" machine semi-capable, but they still haven't achieved the same level as our meat computer with stereoscopic vision in all scenarios.
Anything operating in a real world environment has to be infinitely better than the Roomba bumping around the house or even the dog bouncing off its invisible fence barrier in the yard. The people I talk with like to say the machine needs to have a goal, constantly sense the environment, constantly plan/replan based on the sensing to achieve the goal, act on the current plan, and CONSTANTLY VALIDATE THE OUTCOME to give feedback to the plan in order to achieve the goal. Or even CHANGE the goal. For an old controls engineer like myself (another brief lifetime) it sounds like the most complicated PLC logic and PID loops i can imagine to enact the desired outcomes through a machine.
One may think that something as "simple" as a Waymo vehicle only has a goal to get a rider from point A to B as its outcome. But that's not necessarily the case. What if the rider has a medical emergency in the vehicle and needs to be taken to a hospital? What if an object from the road is trapped under the vehicle and damaging the road? What if a person is approaching the vehicle violently? What if the vehicle has cargo on the roof or truck bed and the cargo is falling out on the road? True autonomy would require the ability to sense all of these needs and react accordingly which may not be just pulling over to the side of road and calling for help. The real world is utterly chaotic at all times and when you have an infinite number of edge case .01% likelihood events, you end up encountering "edge case" at high enough frequency that "autonomy" tends to break down. When you think about the sensing needed to address the above examples, and an infinite number of other scenarios, you can begin to understand why the capabilities need to be resident within the vehicle and why any artificial indicators in the outside world would likely become redundant.
There may be a separate discussion regarding the current state of technology and a potential interim step with artificial outside world indicators due to current technological limitations and safety concerns. I'm not deep enough into it to comment. I will say I'm not sold on Elon's stance that LiDAR is not needed. I may be able to navigate the world just fine without it with my human vision capability, but it seems like there are use cases where it could come in handy in my day to day life. Having been in the manufacturing world for well over two decades, I suspect the downplaying of LiDAR could simply be that the form factor and cost are prohibitive for commercialization. I have no data to back that up other than having seen many features shelved for years until they can be fit into machine forms and profitable.
Jeff
