Tags:#Robotics
ENZHActions Beyond Words: Why LLM Agents Struggle with Contact-Rich ManipulationLLM agents handle many quasi-static tasks with semantics and geometry. Dexterous actions lack clear semantics and must be learned from physical feedback.
6 min read
ENZHHuman Demonstration in Dexterous Manipulation: Semantics and Motion PriorsHumanoids track human motion well; for dexterous hands, contact, embodiment gaps and scale break tracking. Human demos give semantics and coarse motion priors.
7 min read
ENZHHand Motion RetargetingInteraction-aware hand motion retargeting spanning geometry, force, and self-supervision.
6 min read
ENZHReal-World RL: From Single-Task Learning to Last-Mile Training for General PoliciesReal-world RL has moved from proving robots can learn on hardware, to reliable single-task systems, to fixing long-tail failures of general VLA policies.
11 min read