Michael Lutter

About

Michael Lutter is a robotics and machine learning researcher whose work sits at the intersection of physics-based modeling and deep learning, with a particular focus on making autonomous systems more capable, data-efficient, and physically interpretable. He is best known for pioneering **Deep Lagrangian Networks (DeLaN)**, which embed Lagrangian mechanics directly into neural network architectures as a physics-informed prior — work that has accumulated over 250 citations across its iterations and stands as a landmark contribution to physics-informed machine learning for robotics. His research demonstrates that incorporating domain knowledge from classical mechanics dramatically improves generalization and sample efficiency compared to purely black-box approaches. Beyond model learning, Lutter has tackled challenging real-world robotics problems, including high-acceleration reinforcement learning for robot juggling using only binary rewards, and continuous-time dynamics learning through differentiable Newton-Euler algorithms. His work on skill learning systems further reflects a commitment to practical robot deployment in unstructured environments. Across his career, Lutter has consistently bridged the gap between theoretical elegance and real-world applicability, offering the robotics community tools that are both principled and deployable — a combination that makes his research particularly valuable for students and engineers working on next-generation intelligent robotic systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning

   171 citations · 2023
2. 2
   Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning

   81 citations · 2019
3. 3
   Combining physics and deep learning to learn continuous-time dynamics models

   53 citations · 2023
4. 4
   Online Learning of an Open-Ended Skill Library for Collaborative Tasks

   16 citations · 2018
5. 5
   High Acceleration Reinforcement Learning for Real-World Juggling with\n Binary Rewards

   13 citations · 2020
6. 6
   A Differentiable Newton–Euler Algorithm for Real-World Robotics

   6 citations · 2023
7. 7
   Building Skill Learning Systems for Robotics

   5 citations · 2021
8. 8
   Deep Lagrangian Networks for end-to-end learning of energy-based control for under-actuated systems

   5 citations · 2019
9. 9
   High Acceleration Reinforcement Learning for Real-World Juggling with Binary Rewards

   5 citations · 2020
10. 10
    Continuous-Time Fitted Value Iteration for Robust Policies

    5 citations · 2022

Key Collaborators

Contact & Links