Max Robotics — Global Service Robot Directory + US Certification

About

Peter Welinder is a leading researcher at the intersection of reinforcement learning (RL) and robotics, renowned for pioneering techniques that bridge the simulation-to-reality (sim-to-real) gap. His work centers on enabling robots to learn complex, dexterous manipulation skills entirely in simulation before transferring them to physical hardware. Welinder’s most celebrated contribution is the development of **Automatic Domain Randomization (ADR)** , a breakthrough algorithm that allowed a robot hand to solve a Rubik’s Cube—a feat of unprecedented complexity—using only simulated training data (632 citations). He also co-created **Hindsight Experience Replay (HER)** , a seminal algorithm that revolutionized learning from sparse, binary rewards, avoiding the need for intricate reward engineering (352 citations). His 2019 paper on learning dexterous in-hand manipulation via RL, which achieved vision-based object reorientation on a Shadow Dexterous Hand, has garnered over 1,588 citations, underscoring its impact. Welinder has also advanced multi-goal RL with challenging robotics benchmarks and championed sim-to-real transfer through workshops and surveys. As a key figure at OpenAI, his work has set new standards for sample-efficient, scalable robot learning, inspiring a generation of researchers to tackle real-world manipulation with simulated training.