A driverless car called Boss has scooped a $2m prize in a Californian race for robotic vehicles.
Boss successfully drove around an urban environment, avoiding other cars, and covering 60 miles (85km) in less than six hours, all without any human control.
The modified Chevrolet Tahoe was one of six cars that crossed the finish line, from a pack of 11 robotic vehicles which set off at dawn. The others had to pull out after crashes or other problems.
The race was organised by the US military's Defence Advanced Research Projects Agency (Darpa) and is designed to develop unmanned vehicles that could be used in battle situations. Automotive manufacturers say the technology could eventually lead to self-driving cars.
Boss navigated around a simulated town, created on a disused US Air Force base in Victorville, in the Californian desert.
It had to deal with single and dual carriageway roads, junctions, buildings and car parks. As well as the 10 other driverless cars, Boss shared the road with more than 30 professional human drivers to simulate busy traffic.
The robotic vehicle was created by Tartan Racing, a team formed by Carnegie Mellon University in Pittsburgh, Pennsylvania, and General Motors (GM).
"This is a big day for robotics," said Chris Urmson, the team's technology leader.
"We had 11 vehicles that were incredibly capable, and the fact that six of them drove the 60 miles is amazing - just a big day."
Spectators watched from grandstands as the vehicles pulled up at junctions, turned on their indicators, and then pulled away with the steering wheel eerily moving by itself.
Many of them were hoping for some drama, and the rumour on site was the human drivers had been placing bets on who would be hit first.
Each robotic car was followed by a human in a pace car equipped with a kill-switch, designed to stop the robot instantly if it started acting erratically.
Rapid use of the switch avoided any major accidents but there were some minor ones.
A dumper truck called TerraMax was the largest vehicle in the race. At 3m tall, and 2.5 metres wide, the course had to be widened so it would fit.
It drew gasps from the crowd as it pulled into the starting line-up, and bigger gasps a couple of hours later when it became confused as it navigated around the urban environment, and drove into an empty shop.
"It seemed like it was parked, and then all of a sudden it climbed up the curb, and nudged itself into that building," according to Juan, who stood watching from behind a safety barrier.
"It wasn't very fast, but it was definitely hard!"
Big screens around the site repeated images of two other robot cars colliding, but damage was minor enough for them both to continue.
Most of the vehicles look like normal road cars but with an array of sensors mounted on the roof, on each corner and embedded in the bumpers.
One of the key pieces of technology for the winning team was a Lidar - a spinning laser scanner.
"It has 64 individual lasers in it, and it spins about 10 times a second to generate about a million measurements of the world," explained Chris Urmson.
"That gives us a kind of point cloud which we can use to help understand where features are - cars, walls, the sidewalk and so on."
The time it takes for the laser to be fired out and bounce back off an object allows a computer to calculate how far the object is. That information is combined with data from radar sensors for longer-range perception and cameras.
"We have algorithms which process that data and analyse where the fixed objects are, where the moving objects are, and where we are in relation to our maps of the world," Mr Urmson added.
The car industry is watching developments closely.
Larry Burns, GM's vice-president for research and development and strategic planning, said developing cars that drive themselves is a key objective.
"Imagine being able to talk on the phone, eat your breakfast, handle your emails, and leave the driving to the vehicle," he added.
"That would be pretty phenomenal. It's going to a big breakthrough. It's technology that's on the way to 'having cars that don't crash'."
He believes cars with that level of intelligence could be on the road by 2015.