Learning by Watching: A Review of Video-Based Learning Approaches for Robot Manipulation

Robot learning of manipulation skills is hindered by the scarcity of diverse, unbiased datasets. While curated datasets can help, challenges remain regarding generalizability and real-world transfer. Meanwhile, large-scale ‘in-the-wild’ video datasets have driven progress in computer vision using self-supervised techniques. Translating this to robotics, recent works have explored learning manipulation skills using abundant passive videos sourced online. Showing promising results, such video-based learning paradigms provide scalable supervision and reduce dataset bias. This survey reviews foundations such as video feature representation learning techniques, object affordance understanding, 3D hand and body modeling, and large-scale robotic resources, as well as emerging techniques for acquiring robot manipulation skills from uncontrolled video demonstrations. We discuss how learning from only observing large-scale human videos can enhance generalization and sample efficiency for robotic manipulation. The survey summarizes video-based learning approaches, analyzes their benefits over standard datasets, survey metrics and benchmarks, and discusses open challenges and future directions in this nascent domain at the intersection of computer vision, natural language processing, and robot learning.