At the risk of sounding pedantic this is not a Level 4 demo. It is a Level 2 demo where the driver didn't have to intervene during the demonstrated route.
What's the difference? Apart from the legal requirement to have a diver ready to take over in test vehicles (which necessarily makes it Level 2), the fundamental difference is that you'd have to show a lot more than one demo to establish that you've achieved Level 4. Level 4s are supposed to be able to operate without human intervention at all within prescribed domains (e.g. downtown cities). That doesn't mean operate one trip or one day or one month without a disengagement -- that's still Level 2.
I'm super impressed by the demo but Cruise will have to show more data to back up a Level 4 claim.
I agree completely. And the video although not new is quite impressive.
But if the claim of the author that San Francisco is a tough city in which to drive is true, than I fear that some parts of Italy (and probably A LOT of cities in the world) won't have self driving cars for a very long time.
When I saw the video the first time I was very impressed, but I would have never imagined that for someone that was a VERY challenging urban scenario.
From the article:
when we see the streets of San Fransisco conquered, then we know that self-driving is ready to come of age.
Seriously? And if you conquer the street of Naples what will come of age?
If you've ever been to those countries, the first thing you notice as a westerner is that the streets are chaos, traffic rules are barely obeyed, and aggressiveness is required to get anywhere.
Meanwhile, Beijing and Dehli have extreme pollution problems partially caused by cars.
If any place was in need of point to point autonomous ride sharing, it's Asia. The reduction in cars would reduce congestion from the outset. But in the future, protocol guided right of way negotiation could reduce congestion even further.
I tried to flag a taxi in downtown Shanghai for 45 minutes and finally gave into the broken English speaking motorcycle taxi who had been circling the block every five minutes.
Riding bitch through Hangpu to Pudong with a laptop in one arm while lane splitting with aggressive (and otherwise bad drivers) all around was a rather exciting experience.
Even with a skilled pilot, the adrenaline rush was similar to my first time tandem skydiving.
I could imagine the robot car having an easier time in a place with minimal expectation to obey traffic rules. It simply becomes an entirely self-serving entity solving a constraint maximization problem. Namely, get from A to B, don't hit anything, don't get hit.
There is quite complex social interaction going on road. Rules and commands are communicated to participants and there are only seconds to understand them and obey
Something you often see in simulation. But I wounded if these simulations ever use different decision engines and criteria for different vehicles. The different manufacturers are going to produce their own software (some at least) and not treat an encounter the same way. Are we going to have race conditions or "database" lock-like situations (not sure what to call it), as cars try to interpret each others behaviour?
This example, and most simulations, seem to use an algorithm which is very close to "if there is space for me to drive, drive".
In the medium term, I expect self driving cars will instead start to communicate with each other to find a better approximation of optional traffic flow (by finding the solution that is best for the negotiating vehicles). This has several competitive advantages over the simple algorithm: it enhances safety for participating vehicles by removing prediction errors, and if you can negotiate with the other car instead of guessing what it will do you can do much tighter manoeuvers with smaller seaters margins. And of course better traffic flow makes driving more pleasant, which is good for car sales. Given all those market forces I think it's very likely to happen.
In the medium term I agree, but there are som major standards issues and negotiations, like common implementation against the law in several major markets, to work out first, I think.
> "If you've ever been to those countries, the first thing you notice as a westerner is that the streets are chaos, traffic rules are barely obeyed, and aggressiveness is required to get anywhere."
I'm going to get down voted for this, but it has to be said.
The whole "well it couldn't drive in X country" scenario is ridiculous. Driving in unruly, chaotic, unregulated environments is NOT a problem for self driving cars to solve.
That type of driving environment certainly IS a problem, but one that has already been solved by good governance, including laws, rules, and regulations for improving safety of traffic. Countries like the United States, among others, have developed a set of rules and regulations that make self driving cars possible within their domain. It's foolhardy to assume self driving cars can, or should, work outside of that domain.
If you're solution to the problem is to not solve it under the constraints offered, then you're not really solving it. Someone could easily use your argument to say we shouldn't have self driving cars in the US either when we can just use automated trains.
> "If you're solution to the problem is to not solve it under the constraints offered, then you're not really solving it."
That is of course, unless, you DO solve it under other constraints. Sure, you haven't solved X problem with Y solution, but you've still solved X problem. Self driving cars may not be the solution to ALL kinds of traffic problems, but it turns out, they don't have to be.
From my experience in my home country, the only rule of the road there is the survival of the strong. I would imagine that designing algorithms that follow only one rule should actually be much easier. An unpleasant and dangerous environment like this would actually benefit much more from self-driving technology.
There's also the issue of technical problems with for example traffic lights. I remember an occasion in Berlin where the traffic lights broke down on a major intersection with long queues. The right of way ended up being solved in a contest between the status quo competed with the impatience of those who were waiting. The only way as a pedestrian to get across was to step out into the street so cars would stop, which was a bit uncomfortable. Worked out alright, if not terribly efficient. Fixed pretty fast, I think, but an AI would still have to handle, uh, dynamic traffic in western countries.
Ideally, the AI should be able to figure out how the dynamics of such a situation work and not create a jam or something.
In Germany a completely black traffic light has no meaning, rules are as if it wasn't there. Most traffic lights have the usual signs for priority roads for this situation.
But a broken traffic light is a situation that is unfamiliar to drivers, and at peak hours the crossing is guaranteed to be either difficult to navigate or to have suboptimal traffic patterns (otherwise there would be no traffic light in the first place). And of course pedestrians have to improvise (and Germany has a lot of pedestrians compared to the US)
Naples is a bit extreme. Driving there is insane, but even other more normal European cities like London and Paris are a lot harder than any American cities.
I still haven't seen a driverless car handle a single-lane road (e.g. with parked cars on the side). Something like this:
I was very impressed at the number of fringe cases it handled.
A car forcing a merge right in front of it, an armored car taking a lane and a half where it had to wait to go around (there were numerous instances of similar lane blockages that it navigated around), a pedestrian jay walking in front of it, a bus picking up passengers, a bike cutting in front of it, and car pulling out in front of it in heavy traffic.
I had no idea they could handle that many scenarios already.
I wondered that too. Pretty ridiculous that they claim "Self-Driving on the Streets of San Francisco Is Really Hard". City driving in the US is pretty much as easy as it gets. These guys are up for a real challenge when they try to make it work in Delhi, Bangkok et al.
What's the difference? Apart from the legal requirement to have a diver ready to take over in test vehicles (which necessarily makes it Level 2), the fundamental difference is that you'd have to show a lot more than one demo to establish that you've achieved Level 4. Level 4s are supposed to be able to operate without human intervention at all within prescribed domains (e.g. downtown cities). That doesn't mean operate one trip or one day or one month without a disengagement -- that's still Level 2.
I'm super impressed by the demo but Cruise will have to show more data to back up a Level 4 claim.
reply