MIT’s Cheetah 3 robot is effectively blind
Chipmakers like Nvidia and Qualcomm have been busy building products to bring vision intelligence to robots, but in some scenarios, robots may be better off relying on other capabilities to navigate their surroundings.
That’s why the latest version of MIT’s Cheetah robot, the Cheetah 3, is designed to move across rough terrain and through obstacles without relying on vision.
“Vision can be noisy, slightly inaccurate, and sometimes not available, and if you rely too much on vision, your robot has to be very accurate in position and eventually will be slow,” the robot’s designer, MIT Associate Prof. Sangbae Kim, said in a release. “So we want the robot to rely more on tactile information. That way, it can handle unexpected obstacles while moving fast.”
The Cheetah 3 relies on two algorithms: a contact detection algorithm and a model-predictive control algorithm.
The contact detection algorithm helps the robot figure out when it should be swinging its leg and when it should step down. It does so using data from gyroscopes, accelerometers and the joint positions of the Cheetah’s legs. Kim’s team tested this algorithm by having the Cheetah 3 trot on a treadmill and climbing staircase, both of which were covered with random obstacles like wooden blocks.
Meanwhile, the model-predictive control algorithm predicts how much force the Cheetah’s leg should apply once it has committed to a step. The researchers tested this algorithm by kicking and shoving the robot as it trotted on the treadmill, which prompted the robot to produce counter-forces to regain its balance.
The robot, MIT says, would be useful for missions in dangerous or inaccessible environments, such as exploring disaster zones.
While the robot is designed to use “blind locomotion,” the research team has added cameras to the Cheetah 3 so it can give visual feedback of its surroundings. Kim’s team is presenting the robot in October at the International Conference on Intelligent Robots in Madrid.