Not a member yet? Register for full benefits!

Self-Driving AI Vehicles In 2010

This report consists of an overview of the state of progress towards general uptake of self-driving vehicles on civilian roads, world-wide, by February 2010. Several projects are discussed, along with their limitations and degree of separation from AI driving on normal roads.


Heathrow's 'JohnnyCabs', Rome's CyberCars, Castell?n's dual mode busses

Stanford's Stanley AI

Further Reading

Heathrow's 'JohnnyCabs', Rome's CyberCars, Castell?n's dual mode busses

All three of these projects are using the same technology. They come under the European Union's CityMobil project. As the project's own promotional material says:

CityMobil is an Integrated Project, co-funded by the EU, which has been set up to build up knowledge of the issues arising from the integration of automated transport systems in the urban environment. The technical capacity of automated transport vehicles has been tied and tested in previous projects such as Cybercars, Edict and Startdust.

However, knowledge on the wider implications of implementing an automated transport system in a city is not so well established.

This is in no small part because the CityMobil project is taking a fundamentally different paradigm to the US DARPA driven project. Instead of each car being controlled by an AI housed in the car itself, they are drone machines controlled via a central computer system.

Ultimately, this type of system is almost guaranteed to fail because it will not scale up well to a large number of vehicles. However, for small systems - the Heathrow version has 19 vehicles - it works well. Scaling it to 50 million vehicles, would simply not be possible whilst maintaining a secure system.

Heathrow Airport's system was constructed in August 2008, and was the first of the three. The others have also started operation, but all three share the same basic technology: They run on dedicated, custom roads on which no other traffic is permitted to operate.

Heathrow's Automated Cars

Magnetic sensor strips built along the roads at short intervals, networked together under the tarmac, serve as guide strips for the system, telling the central computer system where the cars are on the road system as they pass over them. The vehicles pick passengers up at set points along the route and relay them to other set points. Once a car is in motion, its destination cannot be changed.

They run essentially between terminal 5 and the car-park, with stops in-between. The system in Rome is essentially exactly the same. There, the vehicles are known as CyberCars, and they pick up visitors at a car park or one of a handful of train stations, then take them to a new exhibition centre. Again, specialised roads with no other traffic on them, are required.

In the Spanish city of Castell?n, the driverless busses are a larger version of the same vehicle. Yet again, specialised roads are required with no other traffic, or the system simply does not work.

In all three cases, if there is a person in front of the vehicle, its not up to the vehicle to stop, its up to the centralised computer system to stop the entire system. I.e. all vehicles shut down till the blockage is cleared.

This system has no future in terms of general purpose transportation or even general public transportation. It is a specialised travel gimmick only and does not solve the problem of urban transportation so much as sidestep the whole issue. Its contribution towards the advancement of AI drivers on public roads must therefore be ruled as nil.

Stanford's Stanley AI

In November 2007, the DARPA urban challenge was held at the former George Airforce Base in California. 50 AI controlled independent cars completed, at the same time, to pass the California driver's test on the air base's roads. In amongst them were 50 human-controlled vehicles, just driving around ad deliberately getting in the way. The AI was not told the itinerary ahead of time, and each vehicle drove itself. Three of the AIs passed the challenge, of which one of them, Stanley, created by Stanford University, was the leader.

Stanley in 2007

That challenge proved that AI independent road vehicles could operate alongside human vehicles safely from a technical standpoint, on urban roads. Still, there was some way to go towards wide adoption of them in general society. The sensor systems positioned on the cars, and the large amount of room taken up inside by the computers running the AI, did work against that occurrence. Time, and miniaturisation of both elements, was needed for mass use from a civilian point of view.

After the Urban Challenge, clamour around the vehicles died down. Now, in 2010, it is starting to build up again, as Stanley has re-emerged, attempting to push the envelope for self-driven vehicles just that little further.

Pikes Peak isn Colorado, USA, is a 14,000-foot mountain whose winding road reaches right to the summit. Since 1916 an annual race has been held there, as human drivers push their cars to the limit to try and scramble over them first. This year, one of those drivers isn't human.

The Audi that will attempt Pikes Peak is named Shelley after Michele Mouton, the first woman to win the uphill climb. It is based directly on the Stanley AI system, but with three years worth of improvements. Unlike the original Stanley and Junior, who sense the road with radars and cameras, Shelley will follow a GPS trail not at all dissimilar from the GPS found in TomTom and similar devices, from start to finish. The hope is she will be able to navigate on this information, sufficiently well to be able to use it to later drive on normal roads. Obviously some sensors will be needed to react to changing road conditions, such as wildlife or people stepping out into the road, or another car braking unexpectedly. However, using the GPS for the bulk of actual road assessment greatly reduces the size of the sensor load-out.

"Our first goal is to go up Pikes Peak at speeds resembling race speeds, keep the car stable around the corners and have everything work the way we want it to," said Chris Gerdes, program director of CARS and leader of the graduate research team. "We're not going to put it on the mountain until we can do it safely."

Shelley will attempt a timed race in September when she can get the track to herself.

"Our goal is to show that we can do this," Gerdes said. "There are some sheer drops at Pikes Peak in which any sort of self-preservation kicks in and you slow down a bit. We want to go up at the speed that few normal drivers would ever think of attempting."

The Stanley/Shelley system is also not aimed at immediately replacing human drivers on roads, but is more promising than CityMobil, as the team are still thinking about integrating their AI cars with other road users on real roads, as opposed to specially constructed ones kept apart from anyone else. Still, the upshot of the level of AI capability shown for 2010, remains disappointing in terms of getting AI drivers onto human roads in general.

Further Reading

The JohnnyCab: Total Recall's Look at AI Vehicles

The 2007 DARPA Urban Challenge

Offering The Independence of the Open Road to those to whom Driving is Closed


CityMobil Promotional Material (PDF)

Robot Electric Taxis "Cybercars"

Stanford's robotic Audi to brave Pikes Peak without a driver

Staff Comments


Untitled Document .