The Bayesian Road
From a new traffic-themed issue of Philosophical Transactions of the Royal Society, I was particularly interested in an article by Mark Campbell, et al., “Autonomous driving in urban environments: approaches, lessons and challenges.” As I discuss vis a vis Stanford’s “Junior” in Traffic, the perceptual and interaction dynamics of autonomous driving are infinitely complex. Here’s one bit, which follows a case study in which a vehicle had executed a maneuver that, while not being against the safety rules per se, were “still undesirable” — because, in essence, the vehicle, despite being equipped with formal Bayesian estimators and the like, had failed to take into account for what other vehicles might due. In other words, how do you program a vehicle to “expect the unexpected”?
In order for autonomous driving to reach its full potential, it is vitally important that the cars cooperate in the sense that they agree on traffic rules, whose turn it is to drive through an intersection, and so forth. For this, robust agreement protocols must be developed. Recent work on how to make multiple vehicles agree on common state variables, e.g. using consensus or gossip algorithm (Boyd et al. 2006; Olfati-Saber et al. 2007), provides a promising starting point for this undertaking.
When running such agreement algorithms, it is conceivable that not all vehicles will cooperate. They may, for example, be faulty, or simply driven by human operators, and such vehicles must be identified and isolated in order to balance autonomy with human inputs. This will be true on individual cars, but even more so in mixed human–robot networks. Questions of particular importance (that will have to be resolved using the available interconnections) include the following. (i) Safety: autonomous cars must be able to identify human-driven cars and then not drive into them even though they may violate the robot driving protocol. (ii) Opportunism on behalf of the human drivers: people are already driving badly on the road when the other cars are driven by people. How will they act if no-one is driving? This needs to be taken into account by the autonomous cars (i.e. not only will people not follow the ‘correct’ protocol—they might be outright hostile). (iii) Collaborative versus non-collaborative driving: how should non-cooperative vehicles be handled in an algorithmically safe yet equitable manner?
![[del.icio.us]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/delicious.png)
![[Digg]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/digg.png)
![[Facebook]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/facebook.png)
![[Google]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/google.png)
![[MySpace]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/myspace.png)
![[Slashdot]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/slashdot.png)
![[StumbleUpon]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/stumbleupon.png)
![[Yahoo!]](http://www.howwedrive.com/wp-content/plugins/bookmarkify/yahoo.png)
This entry was posted on Thursday, September 23rd, 2010 at 9:06 am and is filed under Cars, Traffic Wonkery. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.
| 
September 23rd, 2010 at 12:25 pm
I do agree that for the absolutely full potential, you want this level of identification and communication. But it would be a mistake to ask for it at the start. The first robocar on the road has nobody to talk to or identify — everybody else is human driven, probably. You can ask other cars to identify over a radio protocol, and decide how much to trust them, but most cars won’t answer and you have to assume an HDV or unknown robot.
However, one factor left out is that robocars are, by their nature, always capturing the actions of other cars on video, radar and lidar, including any idents like licence plates. And alas, though it is bad news for privacy, almost surely recording this at first to aid in development and debugging. So any infraction will be recorded.
This becomes necessary because, since robots will avoid any impact, they will always lose a game of chicken, and so a Boston driver can just zoom right through a crowded stream of robocars as though they are not there. They will always drive with a small gap they can move into if somebody was about to hit them, and they will use it, parting like fish.
To punish the driver (or robot) playing chicken chances are the video record of this would be used. I don’t like that solution but it is the likely one.
I actually hope for a vast simplification of the vehicle code for robots. The new code would say “Be safe, and don’t unfairly impede others” with occasional local tweaks for specific sections of street. Like the cities that took away all their street signs and rules, the robots will figure it out, and without overt communicating, as a school of fish or swarm of bugs do.
September 23rd, 2010 at 9:31 pm
Or everyone who prefers not to have to drive could just walk, bike, or take the train. Why invest all this effort in the complexification of driving? Gas at $5 and $10 a gallon, coming eventually to a nation near you, will simplify a whole lot of roads and highways at zero cost.
September 24th, 2010 at 11:48 am
I’ve never heard of a “gossip algorithm” but I’m tempted to look it up.
I think all this speculation may be getting a bit over ourselves for two reasons. First, that it’s only been in the last year that anyone has constructed an autonomous vehicle that can navigate an unoccupied training course at low speed, let alone a freeway at rushhour. Second, a lot of recent work in robotics, especially by Rodney Brooks, indicates that the future of robotic intelligence won’t involve ‘programing’ as we understand it at all but rather teaching/learning much like with humans and animals.
In any case, the next ten years should be really exciting.
September 24th, 2010 at 4:22 pm
There’s no mention of bicycles. How will robot cars deal with them? What assumptions about bicycles will be built into the robot software?
September 26th, 2010 at 9:54 am
There was once a person who said, about airplanes, “they will never….” Well, mark my words with this autonomous car thing– it is more fundamental and will not happen– even though it may be forced down our throats at first, vis a vis my most recent entries in Best Driver In The World. There is a freedom of control, each second with a car, that the general public will not want to give up. For example, when I change my mind and decide to go to the store instead of home– with a visual prompt from getting near the store’s region– I can have two choices in the future: reprogram the car, or just mentally decide to go there. Which is the easier? Which system might break down?
No way Jose is this the right future path. If autonomous cars are forced by advertisement and then happen (and Tom, I can certainly see how being in this crowd would benefit your journalistic success), we will then go back to simplicity, where the driver is the key safety system.
John
First Google hit for “best driver in the world”
September 26th, 2010 at 8:47 pm
Indeed, a whole lot of thought is expended on “car vs. car” interactions, but what about the hundred million other things that populate our streets? As a bicyclist, I am horrified. But then I think, robots will never try to ‘teach me a lesson.’ So I don’t know what to think.
September 27th, 2010 at 7:26 pm
No, the programmer will make every car “teach a lesson” to every cyclist. Or, autonomous cars can be restricted to controlled access roads, like Interstates, where bicycles are already prohibited. I see that as the more likely case.
September 28th, 2010 at 7:29 am
Will autonomous cars be programmed to drive the way people actually drive, or the way they’re supposed to drive? For example, where I live the law says drivers have to stop for pedestrians at crosswalks, but studies show less than 10% of drivers actually do.
October 4th, 2010 at 7:54 am
“In order for autonomous driving to reach its full potential, it is vitally important that the cars cooperate in the sense that they agree on traffic rules..” Couldn’t agree more. I posted a blog this morning sort of talking in useful terms about the same concept:
http://amanda-ingle.com/cars/aggressive-driving-why-it-can-be-important/