Max Robotics — Global Service Robot Directory + US Certification

About

Erick Rosete-Beas is a robotics researcher advancing the frontier of autonomous skill acquisition and long-horizon manipulation. His work centers on enabling robots to learn, adapt, and compose complex behaviors from limited data, bridging reinforcement learning, imitation learning, and video prediction. His most influential paper, "Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models" (14 citations), tackles the core challenge of adapting learned skills to noisy real-world perception and dynamics. He contributed to the CALVIN benchmark (6 citations), a foundational tool for evaluating language-conditioned policy learning in long-horizon tasks, helping standardize progress in this critical area. Rosete-Beas also developed "Latent Plans for Task-Agnostic Offline Reinforcement Learning" (5 citations), addressing the difficult problem of learning multi-subtask sequences from offline data without explicit task labels. His work on "T3VIP" (4 citations) introduces transformation-based 3D video prediction, enabling robots to learn physical world dynamics from past experience. Collectively, his research provides key building blocks for general-purpose robots that can interpret language, adapt to changing conditions, and plan over extended horizons—essential capabilities for real-world deployment.