Robot Motion Generation

Robot motion generation studies how to turn goals, observations, language, or demonstrations into trajectories that robots can execute. Approaches range from optimization, model-predictive control, splines, and Dynamic Movement Primitives to learned policies using diffusion, flow matching, and video or vision-language models; many combine these with reinforcement learning or motion retargeting. Work spans manipulation, humanoid locomotion, navigation, and human-robot interaction, with evaluation on benchmarks such as LIBERO and CALVIN as well as real robots. A central challenge is balancing diverse, expressive motions and fast generation with physical feasibility, safety, smooth closed-loop execution, and transfer across robots and changing environments.