… Under construction, text, images and videos coming soon …

I work on inductive biases for robot learning. Within this field I focus on whether inductive biases are preferable for learning algorithms and how inductive biases can be incorporated in generic deep learning algorithms. I showed that (1) inductive biases can drastically improve sample efficiency for robot learning and (2) that generic inductive biases can be incorporated in deep learning models to obtain interpretable and physically plausible models.

A full list of all publications can be found at Google Scholar

Deep Lagrangian Networks (DeLaN)

Paper:

  • Lutter, M.; Ritter, C.; Peters, J. (2019). Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning, International Conference on Learning Representations (ICLR). [Arxiv]

  • Lutter, M.; Peters, J. (2019). Deep Lagrangian Networks for end-to-end learning of energy-based control for under-actuated systems, International Conference on Intelligent Robots and Systems (IROS). [Arxiv] [Code]

Continuous Time Optimal Feedback Control

Paper:

  • Lutter, M.; Belousov, B.; Listmann, K.; Clever, D.; Peters, J. (2019). HJB Optimal Feedback Control with Deep Differential Value Functions and Action Constraints, Conference on Robot Learning (CoRL) [Arxiv]

Differentiable Newton-Euler-Algorithms

Paper:

  • Lutter, M.*; Silberbauer, J.*; Watson, J.; Peters, J. (2020). A Differentiable Newton Euler Algorithm for Multi-body Model Learning, ICML Workshop on Inductive Biases, Invariances and Generalization in RL [Arxiv]

  • Lutter, M.*; Silberbauer, J.*; Watson, J.; Peters, J. (2020). Differentiable Physics Models for Real-world Offline Model-based Reinforcement Learning, [Arxiv] [Videos]

Robot Juggling

Paper:

  • Ploeger, K.*; Lutter, M.*; Peters, J. (2020). High Acceleration Reinforcement Learning for Real-World Juggling with Binary Rewards, Conference on Robot Learning (CoRL). [Arxiv] [Videos]