Max Robotics — Global Service Robot Directory + US Certification

About

Raul Sampedro is a leading researcher in robotics and artificial intelligence, specializing in scalable methods for training generalist agents. His work bridges the gap between large-scale, unsupervised data and complex sequential decision-making, with a focus on robotic manipulation and learning from video. Sampedro’s most influential contribution is the **Video PreTraining (VPT)** framework, which demonstrates how models can learn to act by watching unlabeled online videos, achieving 50 citations and setting a new paradigm for pretraining in domains like robotics and computer use. He also pioneered **asymmetric self-play for automatic goal discovery**, a method where two agents (Alice and Bob) generate and solve increasingly challenging tasks, enabling a single policy to handle unseen goals and objects in robotic manipulation. This work, with 21 citations, showcases his ability to combine game theory with practical robotics. Sampedro’s research is notable for its focus on data efficiency and generalization, making him a key figure in advancing embodied AI. His achievements include developing algorithms that reduce the need for human supervision, pushing toward autonomous skill acquisition in real-world environments.