Max Robotics — Global Service Robot Directory + US Certification

About

Bob McGrew is a leading figure in reinforcement learning (RL) for robotics, best known for pioneering work that bridges the simulation-to-reality gap. His research centers on dexterous manipulation, sparse-reward exploration, and domain randomization. McGrew co-authored the landmark paper "Learning dexterous in-hand manipulation" (1,588 citations), which used RL to train a Shadow Dexterous Hand to reorient objects in the real world—a feat previously considered intractable. He further demonstrated the power of simulation-trained policies by solving a Rubik's Cube with a robot hand (632 citations), introducing automatic domain randomization (ADR) to enable zero-shot transfer to physical hardware. McGrew also contributed foundational algorithms like Hindsight Experience Replay (352 citations), which revolutionized learning from sparse, binary rewards, and co-created the Fetch robotics environments (196 citations) that became standard benchmarks for multi-goal RL. His work on overcoming exploration with demonstrations (63 citations) and generative models for grasping (24 citations) has shaped modern robotic learning. McGrew’s achievements exemplify how simulated training can unlock real-world robotic capabilities, making him a key innovator in the field.