Robots That Learn from Imagined Worlds
What if robots could learn new tasks just by watching AI-generated videos - and still obey physics?
PhysWorld is a new framework that turns text-and-image prompts into physically executable robot skills.
- It first generates a task-conditioned video from a single scene image and a command.
- Then it reconstructs the scene's physical world - objects and dynamics - from those frames.
- Finally, it grounds the video motions into real actions with object-centric residual reinforcement learning, so the robot follows physics, not just pixels.
Why it matters: this synergy cuts out real robot data collection, enables zero-shot manipulation across diverse tasks, and substantially boosts accuracy versus prior "retarget video to robot" baselines.
Paper: http://arxiv.org/abs/2511.07416v1 • Project: https://pointscoder.github.io/PhysWorld_Web/
Paper: http://arxiv.org/abs/2511.07416v1
Register: https://www.AiFeta.com
Robotics AI ComputerVision ReinforcementLearning GenerativeAI VideoGeneration EmbodiedAI
