This Is How Google's New $1 Million Robot Sees The World

A team of roboticists at MIT is preparing Atlas, a 6-foot-2-inch-tall, 330-pound humanoid robot that can climb stairs, open doors, and even drive a car, for competition in the DARPA Robotics Challenge at the end of 2014.

Atlas is built by Boston Dynamics, the robotics company bought by Google last December, and there are only eight of them in the world.

The DARPA Challenge is a "robot Olympics" - robots are asked to do things like climb a ladder, or connect a firehose to a standpipe and turn on the valve. These tasks are designed to test a robotic system's "mobility, manipulation, dexterity, perception, and operator control mechanisms."

A robot that can't make sense of its world in real time would never be able to do these things. Here's how Atlas' sensor systems rise to the challenge.

In our demo, Atlas was given instructions to find the piece of wood and pick it up.

This is the operator's view, where you give the robot its instructions. A number of monitors provide feedback on what Atlas can see and detect.

Here's what Atlas sees from the spinning "lidar" unit in its head. Lidar (that's light detection and ranging) is an imaging system that makes sense of an environment by bouncing lasers off it and analyzing how they reflect back. It's essentially a very precise rangefinder that operates in three dimensions.

Note the yellow and green "footprints" on the screen - this is Atlas determining where to place its feet in order to approach that steel beam:

Atlas has two stereo proximity cameras ("for situational awareness") and a third camera in its head angled slightly downward. This is a composite image of what the robot "sees" by way of its three cameras.

These systems come together to make it possible for the robot to perform environment-dependent tasks, like identifying the piece of wood it was told to pick up and stooping down to grab it.

Success!