DARPA unTerminators gather for Robotics Challenge finals in Hell*

* Aka Los Angeles


Schwarzenegger-sized robots that can break through walls are the staple of Hollywood, but the real thing is coming to California.

The US military's secret-squirrel super-science-project unit DARPA is preparing to hold the finals of its Robotics Challenge contest in Los Angeles this weekend.

25 teams are flying in to take part in the two-day finals, which will see a humanoid robot take part in a simulated rescue operation. The mechanoids will have to drive a car to the test site, open a door, cross rubble, go up some stairs, cut through a wall, open an industrial rotary valve, and perform one mystery task – all without continuous human control.

Those worried about an army of American killer robots can rest easy, the research wing of the US military says, because the purpose of the competition is to save lives, not take them. The competition is to build an autonomous robot that can enter hazardous areas, like the Fukushima nuclear station, remove injured humans, shut down industrial processes, and clear the area of debris.

Doing this means using equipment designed for humans, so the competitors have to be roughly humanoid. Six of the 25 teams will be using the same 6ft robot – dubbed ATLAS – that's built by Google-owned Boston Dynamics, which you may remember from such other robots as Cheetah, Wildcat, and BigDog.

Google was the hot favorite to win this weekend's contest; its SCHAFT robot won the trials and eschewed DARPA funding. But the web giant's team has pulled out of the contest, reportedly because it doesn't dig the military funding of the project, despite its humanitarian theme.

There has been no further response from the Chocolate Factory on the matter, and Google has been taking flak on the matter. Last year some loon disrupted its I/O developer conference protesting about "killer robots," which may explain the over-the-top security at this year's event.


A few teams have built their own humanoid designs, although some have created four-legged contraptions that can still use human tools, but are a lot more stable on the ground. Standing on two feet is easy when you've had thousands of years of evolutionary experience, but as any roboticist will tell you, it's inefficient and costs a hell of a lot of processing power to manage.

This was reflecting in the trial sessions; robots broke legs, stumbled and fell in rubble, and generally displayed all the cock-ups you'd expect in a first-time competition like this. For this weekend's challenge, DARPA has made the rules even tougher – robots must rely on batteries only; safety tethers to stop them falling over are banned; and the organizers will degrade communications with team controllers so that the robots have to act autonomously.

Given the difficulties involved, don't expect any of the teams to fully complete the finals test – especially since the organizers have one unknown task for the robot entrants, but that's not a bad thing. We saw the same pattern when DARPA held its first autonomous vehicles contest, but ten years later computer-controlled cars have clocked up over a million miles on US roads without an accident.

DARPA is adept at playing the long game with technology development, and so far the results have been pretty good. It developed ARPANET, the network that formed the basis for the internet; has been behind the development of simulation software and drone technology; and helped create the Tor online privacy system.

As for fears that these robots could be used in war, there's not much to worry about at present. They are slow, unwieldy, move at a pace a toddler could outrun, and have a one-hour lifespan between recharging. Given the goals of the challenge, they are not destined to be killers and more likely to say "come with me if you want to live." ®

Sponsored: Minds Mastering Machines - Call for papers now open

Biting the hand that feeds IT © 1998–2018