🤖 Robots Learn Parkour
Scientists from KAIST and Jemin Hwangbo's team demonstrated how to teach the quadruped Raibo to autonomously run through complex, discrete environments, perform jumps, and navigate inclined surfaces. Their system builds an optimal step plan through a hierarchical scheme: first, support points are selected using a neural network and heuristics, then the best route is verified in a physics simulation. As a result, the robot learns to overcome more complex obstacles than before - including running up vertical walls, jumping over 1.3-meter gaps, and moving over rocks at a speed of 4 m/s.
📝 Paper: https://www.science.org/doi/10.1126/scirobotics.ads6192
#ai #agi #robotics