Robot dog training

A robot dog trained entirely on dog's video 🐶 Robots can dance and sprint through forests. But teaching something new, needs millions of trial-and-error examples. I thought: what if a quadruped robot could learn behaviours directly from a real dog, using nothing but a phone video? A month later, it worked. The pipeline in a nutshell: extract the dog's 3D skeleton from a single monocular video, retarget the motion to the robot's joints, solve the IK, then train an RL policy in simulation with full physics until the robot moves like the dog it watched. Everything is open source, and the article expands on these ideas 👇 https://lnkd.in/eswYTG9D #robotics #reinforcementlearning #simulation #computervision #isaacsim