Milo Carroll

Papers

Total Citations

H-Index

About

Milo Carroll is a roboticist advancing the frontier of agile and versatile robot locomotion. His primary research focuses on developing learning-based control frameworks that enable legged robots to navigate complex, unstructured environments with unprecedented adaptability. Carroll’s most notable contribution is his work on kernel-based residual learning, where he pioneered a hybrid approach that combines the stability of model predictive control with the flexibility of reinforcement learning. In his highly cited 2023 paper, he introduced a framework that first trains a kernel neural network on data from an MPC controller, then freezes it while a residual network learns to handle edge cases and dynamic maneuvers through RL. This method allows quadrupedal robots to achieve both robust, efficient gaits and the ability to recover from disturbances or traverse rough terrain. While still early in his career, with over 2 citations on this foundational work, Carroll’s approach has already influenced subsequent research in sim-to-real transfer and adaptive locomotion. His work represents a significant step toward robots that can operate reliably in the real world, bridging the gap between classical control and data-driven methods.