Paul Gesel

Papers

Total Citations

H-Index

About

Paul Gesel is a robotics researcher whose work focuses on advancing **learning from demonstration (LfD)** and **imitation learning (IL)** — key areas that enable robots to acquire skills by observing human teachers. His major contributions lie in making these learning frameworks more robust and practical for real-world deployment. In his most-cited work, "Robust Behavior Cloning with Adversarial Demonstration Detection" (2021, 5 citations), Gesel tackles a critical limitation of standard IL: the unrealistic assumption that all human demonstrations are correct. He proposes a method to detect and handle faulty or adversarial demonstrations, significantly improving a robot's reliability in noisy environments. Earlier, in "Learning Motion Trajectories from Phase Space Analysis of the Demonstration" (2019, 4 citations), he introduced a novel approach to generalize tasks from just a single demonstration by reconstructing motion trajectories using phase space analysis — a technique inspired by dynamical systems theory. His 2020 paper further extends this work by formulating trajectory learning as an optimal control problem, drawing on human movement literature to teach robots activities of daily living. Though early in his career, Gesel’s focus on robustness and sample efficiency addresses foundational challenges in LfD, making his work highly relevant for researchers aiming to deploy robots in unstructured human environments.