Skye Thompson

Massachusetts Institute of Technology, Brown University

Papers

Total Citations

H-Index

About

Skye Thompson is a roboticist advancing the frontier of manipulation through structure, efficiency, and generalization. Their research centers on three interconnected themes: distributed manipulation, structured policy learning for contact-rich tasks, and few-shot imitation learning. Thompson’s work on delta-manipulator arrays (7 citations) introduced a robust, scalable approach to planar translations using cooperative actuator grids, demonstrating how distributed systems can achieve precise, fault-tolerant manipulation. Building on this, they developed Composable Interaction Primitives (CIPs) (4 citations), a policy class that exploits the geometric and temporal structure of sustained-contact skills—such as opening drawers or shifting gears—to dramatically reduce what must be learned from data. Most notably, Thompson’s Interaction Warping method (3 citations) achieves one-shot imitation learning of SE(3) manipulation policies by inferring 3D object meshes from a single demonstration, a breakthrough for open-ended robotic learning. Their work consistently bridges theory and practice, offering principled frameworks that minimize data requirements while maximizing generalization. With each contribution, Thompson is shaping a future where robots learn complex, contact-rich skills as intuitively as humans do.