Kevin Ellis

Papers

Total Citations

H-Index

About

Kevin Ellis is a leading researcher at the intersection of robotics, embodied AI, and neuro-symbolic reasoning. His work focuses on enabling robots to build and leverage abstract world models for more intelligent, generalizable manipulation. A key contribution is the development of **Rapid Motor Adaptation** techniques, which allow robotic manipulator arms to generalize across diverse task conditions—including variations in object shape, density, and friction, as well as external disturbances—without retraining. This highly influential work has already garnered 6 citations. Ellis also introduced **VisualPredicator**, a framework that learns abstract, neuro-symbolic predicates to form task-specific world models. By combining the structure of symbolic logic with the flexibility of neural networks, this approach enables robots to selectively expose essential task elements while ignoring raw sensorimotor complexity, achieving 3 citations. Through these contributions, Ellis is pioneering a path toward robots that can reason, plan, and adapt with human-like efficiency, making him a rising figure in the quest for broadly intelligent embodied agents.