Research & Teaching

Research Interests

馃晩 Swarm Reinforcement Learning
馃搻 Graph Neural Networks and Geometric Deep Learning
馃 (Versatile) Reinforcement Learning from Human Feedback

Publications

Physics-informed MeshGraphNets (PI-MGNs):
Neural finite element solvers for non-stationary and nonlinear simulations on arbitrary meshes

Computer Methods in Applied Mechanics and Engineering (CMAME), 2024.
T. W眉rth, N. Freymuth, C. Zimmerling, G. Neumann, L. K盲rger

Movement Primitive Diffusion: Learning Gentle Robotic Manipulation of Deformable Objects.
IEEE Robotics and Automation Letters (RA鈥慙), 2024.
Scheikl, P. M., Schreiber, N., Haas, C., Freymuth, N., Neumann, G., Lioutikov, R., & Mathis-Ullrich, F.

Swarm Reinforcement Learning for Adaptive Mesh Refinement.
36th Neural Information Processing Systems (NeurIPS), 2023.
N. Freymuth, P. Dahlinger, T. W眉rth, S. Reisch, L. K盲rger, G. Neumann

Grounding Graph Network Simulators using Physical Sensor Observations.
11th International Conference on Learning Representations (ICLR), 2023.
J. Linkerh盲gner, N. Freymuth, P.M. Scheikl, F. Mathis-Ullrich, G. Neumann

Caption Figure for Grounding Graph Network Simulators
Caption Figure for Adversarial Imitation Learning with Preferences

Adversarial Imitation Learning with Preferences.
11th International Conference on Learning Representations (ICLR), 2023.
A. Taranovic, A.G. Kupcsik, N. Freymuth, G. Neumann

Inferring Versatile Behavior from Demonstrations by Matching Geometric Descriptors.
6th Conference on Robot Learning (CoRL), 2022.
N. Freymuth, N. Schreiber, P. Becker, A. Taranovic, G. Neumann

Caption Figure for Versatile Imitation from Geometrically Observed Representation

Workshops

KalMamba: Towards Efficient Probabilistic State Space Models for RL under Uncertainty.
“Next Generation of Sequence Modeling Architectures” and “Aligning Reinforcement Learning Experimentalists and Theorists” Workshops at International Conference on Machine Learning (ICML), 2024. P. Becker, N. Freymuth, G. Neumann

Iterative Sizing Field Prediction for Adaptive Mesh Generation From Expert Demonstrations.
AI4Science Workshop at International Conference on Machine Learning (ICML), 2024.
N. Freymuth, P. Dahlinger, T. W眉rth, P. Becker, A. Taranovic, O. Gr枚nheim, L. K盲rger, G. Neumann

Latent Task-Specific Graph Network Simulators.
AI4Science Workshop at Neural Information Processing Systems (NeurIPS), 2023.
P. Dahlinger, N. Freymuth, M. Volpp, T. Hoang, G. Neumann

Swarm Reinforcement Learning for Adaptive Mesh Refinement (v1). Workshop on Physics for Machine Learning at International Conference on Learning Representations (ICLR), 2023.
N. Freymuth, P. Dahlinger, T. W眉rth, L. K盲rger, G. Neumann

Versatile inverse reinforcement learning via cumulative rewards. 4th Robot Learning Workshop at Neural Information Processing Systems (NeurIPS), 2021.
N. Freymuth, P. Becker, and G. Neumann

Caption Figure for Versatile Inverse Reinforcement Learning

Theses Supervision and Teaching

Bachelor’s Theses

  • Autonomous Partitioning of Heterogeneous Swarm Systems
  • Episodic Versatile Imitation Learning via Geometric Feature Matching
  • Representation-Invariant Latent Spaces for Imitation Learning from Observation
  • Swarm Reinforcement Learning with Graph Neural Networks

Teaching

  • Teaching Assistant for the “Reinforcement Learning” Lecture @ KIT
  • Seminar and Praktikum “Robot Learning” @ KIT

Misc. Projects

  • Efficient Swarm Reinforcement Learning from Human Feedback
  • Spatial Credit Assignment for Swarm Reinforcement Learning

Master’s Theses

  • Adaptive Mesh HP-Refinement with Swarm Reinforcement Learning
  • Explainable Graph Neural Networks for Autonomous Driving
  • Explainable Graph Neural Networks for Traffic Routing
  • Graph Diffusion for Adaptive Mesh R-Refinement
  • Adaptive Mesh Refinement from Human Feedback
  • Variational Swarm Reinforcement Learning for Predictive Models of Deformable Objects
  • Reinforcement Learning for Deformable Object Manipulation
  • Swarm Reinforcement Learning in Limited Visibility with Graph Neural Networks
  • Physical Simulation from Observation using Graph Neural Networks
  • RippleNet: A remote message passing method for Graph Neural Networks