Add to Favourites
To login click here
The Dynamic Robotics Lab at Oregon State University used a machine learning approach to optimize the speed of a bipedal robot, Cassie. With a $1 million grant from DARPA, the team trained Cassie to run the 100m dash in 24.73 seconds, breaking the record. The team used reinforcement learning to train Cassie, using proprioceptive control input from the robot’s joint angles and velocity. This approach allowed the team to optimize the robot’s performance without having to manually work out equations and solve optimization problems.