Contact-Rich Manipulation

Contact-rich manipulation concerns tasks such as pushing, insertion, juggling, and rearranging objects, where friction, impacts, and changing contacts determine the outcome. Approaches combine contact models and trajectory optimization with model-predictive control, hierarchical reinforcement learning, geometric planning, and tactile or visual prediction. Differentiable simulators and domain randomization help plan under uncertain dynamics and transfer policies from simulation to real robots, while compliant control addresses safety and external forces. A central challenge is balancing physically accurate contact reasoning with fast, reliable planning across changing objects, environments, and contact modes.