Max Robotics — Global Service Robot Directory + US Certification

About

Patrick Miller is a leading researcher in robot learning, with a focus on imitation learning, manipulation, and policy evaluation. His most impactful contribution is the DROID dataset (2024, 108 citations), a large-scale, in-the-wild robot manipulation dataset that has become a foundational resource for training and benchmarking robotic policies in real-world, unstructured environments. This work addresses a critical bottleneck in robotics—the lack of diverse, high-quality demonstration data—and has accelerated progress in dexterous manipulation and generative modeling for robotics. Miller also advances the safety and reliability of learned policies. In his 2025 work on uncertainty-aware runtime failure detection, he proposes methods to identify policy failures without requiring explicit failure data, a crucial step for deploying imitation learning in long-horizon, high-stakes tasks. His research on policy comparison with near-optimal stopping further tackles the practical challenge of rigorously evaluating and comparing robot learning methods, ensuring that new approaches are validated against baselines with statistical confidence. By combining large-scale data collection with rigorous evaluation and safety mechanisms, Miller is shaping the future of robust, real-world robotic manipulation.