Robot Manipulation Benchmarks
Overview
They range from simulated tabletop grasping and placement suites such as LIBERO and RoboCasa to real-robot tests of insertion, dexterous hands, bimanual work, mobile manipulation, and deformable or industrial objects. Common designs pair standardized task protocols and demonstrations with controlled shifts in objects, scenes, cameras, dynamics, or instructions; evaluation increasingly adds stage-level progress, recovery, motion quality, force and tactile interaction, and temporal safety to task success. Researchers use simulation, real-world rollouts, and matched sim-to-real protocols to compare imitation- and reinforcement-learning policies, vision-language-action models, and planning or memory systems. The central challenge is making scores both reproducible and diagnostic: success on fixed tasks may hide shortcuts, unsafe behavior, weak generalization, and gaps between simulated and physical performance.
Latest papers
316 papers85 papers in the last four weeks, up 81% on the four weeks before. 0.5% of all new papers.