Papers
5
Total Citations
61
H-Index
4
About
Steve Heim is a roboticist whose research lies at the intersection of legged locomotion, control theory, and reinforcement learning. His work is unified by a central question: how can we design robots and controllers that are both simple and capable? Heim’s most cited paper, "On designing an active tail for legged robots" (24 citations), introduces a design principle that decouples control objectives—such as energy injection and body-pitch stabilization—using an actively controlled tail. This insight simplifies control for steady-state running, offering a practical path to more agile legged robots. Heim is also a leading voice in applying reinforcement learning directly to hardware. His 2023 paper "Benchmarking Potential Based Rewards for Learning Humanoid Locomotion" (21 citations) addresses the critical challenge of reward function design, showing that well-structured shaping rewards can dramatically accelerate learning. In "Shaping in Practice: Training Wheels to Learn Fast Hopping Directly in Hardware" (2017), he demonstrated that robots can learn complex behaviors like fast hopping without simulation, using carefully designed reward structures to guide exploration. More recently, his work on Model Hierarchy Predictive Control (2023) tackles the computational bottleneck of MPC by scheduling simpler models for different planning horizons. Through these contributions, Heim has advanced both the theory and practice of robot learning and control, making complex behaviors more accessible and robust.
Research Focus
Key Achievements
Top Papers
- 1
- 2Benchmarking Potential Based Rewards for Learning Humanoid Locomotion21 citations · 2023
- 3
- 4
- 5