Dynamic Robotic Manipulation

Dynamic robotic manipulation concerns tasks where objects, people, or the robot itself move during contact, such as catching, throwing, handling ropes and cloth, and inserting parts into moving targets. Approaches combine temporal vision and motion prediction with model predictive control, reinforcement learning, and newer diffusion-policy, vision-language-action, and world-model methods; force and tactile feedback can help correct actions during contact. Evaluation increasingly tests real-time performance in simulation and on hardware, including moving-target and deformable-object benchmarks, with attention to latency, changing object properties, and transfer from simulation to real robots. A central challenge is balancing fast, reliable reactions with accurate prediction and generalization, especially when contact dynamics are uncertain and repeated outcomes vary.