a team of researchers at oregon state university have taken on the challenge of teaching a blind bipedal robot to climb stairs. while four-legged bots have long used computer vision to navigate a set of steps, they are not always offered a perfect environment. with such possible conditions as dim lighting or fog, movement can be challenging and robots can not rely on cameras or other sensors. to sidestep these issues, agility robotics’ bipedal robot ‘cassie’ is trained to navigate through ‘proprioception,’ or body awareness. cassie was last put to the test in 2018, walking through fire and riding a segway. after accomplishing such extreme feats, climbing a flight of stairs had proven to be the robot’s boldest achievement yet.
before ‘cassie’ ventured up a flight of physical stairs, the team at oregon state university’s dynamic robotics laboratory trained the blind bipedal robot to learn virtually. this process involved a technique called sim-to-real reinforcement learning (RL) which virtually established how cassie will walk. the team explains: ‘for biped locomotion, the training will involve many falls and crashes, especially early in training.’ to avoid this, the simulator allows the robot to train without damaging itself. after the successful physical tests, the team notes: ‘to our knowledge, this is the first controller for a bipedal, human-scale robot capable of reliably traversing a variety of real-world stairs and other stair-like disturbances using only proprioception.’
dynamic robotics laboratory, oregon state university