MIT’s Cheetah 3 robot doesn’t need sight to navigate stairs

When we first met the Cheetah, a four-legged robot built by engineers at the Massachusetts Institute of Technology (MIT), the machine was, frankly, not all that fascinating. Sure, it could run pretty quickly for a robot. But at 10 mph, the Cheetah was hardly as impressive as Boston Dynamics’ line of robo-dogs, nor could it keep up with its living, feline counterpart.

Four years on, the Cheetah has made some progress thanks to its MIT engineers. Now dubbed the Cheetah 3, the robot’s current version can leap onto tables, handle rough terrain, and even use blind locomotion to navigate. By developing the machine to get around without the use of cameras, the engineers hope to create a robot that can “feel” its way through a room, no matter how dark an environment may be. In the real world, this ability could make the robot suited for reconnaissance and search and rescue missions.

“Blind locomotion is [locomotion] without vision,” Sangbae Kim, a mechanical engineer at MIT and the robot’s designer, told Digital Trends. Vision-oriented movement obviously seems natural to most humans, but it’s data intensive and noisy for machines. In comparison, movement can also be oriented by proprioception (a sense of one’s own body in relation to the world) and vestibular organs (those which help maintain balance). “Most research robots rely too much on vision and don’t utilize the other two enough,” Kim said. “Before we integrate the vision, we want to have robust behaviors first.”

The Cheetah 3 needs some help from humans to navigate blindly (it relies on manual commands for direction and speed) but it’s capable of tackling obstacles, such as stairs, autonomously. It does so by using sensors and algorithms to orient its body to its environment. For example, the robot uses a contact detection algorithm (which helps the robot determine when to move a particular leg) and a model-predictive control algorithm (which helps predict how much force the robot should apply to a given leg).

The Cheetah 3 weighs in at about 90 pounds, with four legs, a determined gait, and exposed wires and circuity. Each of its knee joints are invertible, meaning they can flex and bend in the opposite direction, which lets the robot adjust its stance to gain better balance and perform other double-jointed tricks. Among its new tricks, the Cheetah 3 can also jump onto a 30-inch desk, recover from being pushed, and twist.

In the near term, Kim and his colleagues are developing the Cheetah for use in disaster relief situations or for tasks that are difficult or dangerous for humans to perform. They plan to add an arm to allow the robot to manipulate objects around it.

The researchers will demonstrate their robot in October at the International Conference on Intelligent Robots in Madrid, Spain.