There are a few controverial bits that I'd love to hear what HN thinks.
1. Boston Dynamics Spot, where you have a technical achievement without a clear application
2. What it takes to win in self driving. At one point Musk said "If you have accurate vector space representation, then you're kind of like a video game." That's just not true.
This is actually my entire strategy for approaching autonomy, I make as much of the problem space a game; just a serious one. And the most successful optimizations and solutions are the ones that adopt techniques from games. Many academic solutions are formalizations for old-school game "hacks".
In my line of work, training, in many occasions, is about making deep computer vision match ground truth, generated by game tech.
In many ways, “perception” is solved or close to it. But this is just a subset of the self-driving problem
Think about seeing a playground ball rolling into the street. Most human drivers would anticipate a child jumping into the street after the ball even though they’ve never experienced that situation before because they can correlate the perception with non-driving scenarios (I.e. playground + ball is associated with children). A self-driving car may not have the non-driving context and I’m not sure society is willing to accept the outcome while the cars gather enough data to learn.
And that’s to say nothing of all the off-nominal maintenance related conditions to account for.
Maybe there can be enough data sharing to update models effectively. It will be interesting to see this industry evolve
If you do not allow for frequent doubt, aka re-classification, you get hallucinating systems, imagining a bike-rishka as a slow car with people hanging out
I believe a case of a pedestrian being killed by a self driving car was related to the system reclassifying enough to initiate a deliberate delay timer. In a system traveling 60mph this delay in decision was enough to cause a mishap
The serious accidents involving self-driving cars have all been huge failures at the basic "don't run into things" task. Google/Waymo, which seems to be past that, has subtle problems, like "projected that bus couldn't fit through space to left of car in wide lane and so turned slightly into path of bus and was scraped." The standard Waymo accident is "advanced into intersection, detected cross traffic, stopped, was rear-ended by human driver at low speed".[1]
[1] https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...
It appears the mitigation for a high false positive (I.e., nuisance alarm) was to program in a delay (I.e., “action suppression”) which inadvertently introduced a worse hazard, the timetable from it is in the link below
[1]https://www.theregister.co.uk/2019/11/06/uber_self_driving_c...
- perception: sensors -> vector space (a list of objects with their locations)
- planning: vector space -> path
Both are hard, but for the planning problem you can do most of the training and testing in simulation. It's fairly easy to simulate the response of the car.
The perception problem is more expensive to solve, because you have to collect so much data under every weather and lighting condition.
I met the designer of the Furby and found out what its manufacturing cost is. Let's just say it's less than a meal at most fast food places.
Big Dog and the Legged Squad Support System were impressive, but not useful. The USMC rejected the LS3 and bought small ATVs instead. Much more practical.
(Reminds me of the U.S. Army's flying saucer, the AvroCar. That was a disk-shaped aircraft that was supposed to be a flying Jeep. It was unstable, underpowered, and couldn't get out of its own ground effect. Really cool, though. The Army had another project at the same time, the Utility Helicopter program. That resulted in the UH-1 "Huey", one of the most successful helicopters of all time, with over 16,000 made and many still in use after six decades. It's not exotic but usually will get you there and back.)
I could easily see 1000 qty sales of SpotMini for the cool factor alone -- i.e. research labs, corporate labs, startups, etc without ever finding a real "value" application. This drives me nuts as a roboticist & CEO of a service robotics company. But so it goes -- at least it's better than their past product: Youtube Videos.
What the world needs is a good US$5000 assembly robot. Universal Robots has a good one for $23,000. Nobody is building good robots in high volume. Not even Foxconn, which tried.