The Developments and Challenges towards Dexterous and Embodied Robotic Manipulation: A Survey
Authors: Gaofeng Li, Ruize Wang, Peisen Xu, Qi Ye, Jiming Chen Submitted: July 2025 (v2: November 2025) Venue: arXiv 2507.11840
Abstract
Surveys the evolution of robotic manipulation systems progressing from mechanical programming to embodied intelligence, alongside advances in gripper technology. Focuses on contemporary embodied dexterous manipulation, with emphasis on two primary research directions:
- Data collection approaches: simulation, human demonstrations, teleoperation
- Skill-learning methodologies: imitation and reinforcement learning
Identifies and discusses three fundamental obstacles currently limiting progress in dexterous robotic manipulation.
Structure
- Historical progression: mechanical → programmed → learning-based → embodied
- Gripper evolution: parallel jaw → multi-finger → dexterous hands
- Data collection: sim-to-real, kinesthetic teaching, VR teleoperation, mocap
- Policy learning: BC, GAIL, diffusion, foundation models (VLA)