Mahmoud Selim
Stockholm, Sweden

Mahmoud Selim

Industrial Ph.D. student

TRATON GroupKTH Royal Institute of Technology

EmailGoogle ScholarLinkedIn

I am an industrial Ph.D. student at TRATON Group and the Division of Decision and Control Systems, School of Electrical Engineering and Computer Science, at KTH Royal Institute of Technology. I am advised by Karl H. Johansson.

My research focuses on learning and intelligent behavior in autonomous systems, with an emphasis on generalization, adaptation, and acquiring new behaviors. My current work spans reinforcement learning, adaptive dynamics modeling, and motion planning, with robotics and autonomous driving as real-world applications.

Research interests

  • Reinforcement learning
  • Generalization & adaptation
  • Robot intelligence

My broader goal is to build intelligent agents that keep learning and turn that knowledge into capable behavior in the physical world.

News

Selected publications

All publications on Scholar
ICDP illustrative placeholder

Beyond Policy Support: Interaction-Constrained Offline Reinforcement Learning for Autonomous Driving

Mahmoud Selim, Cristina Cipriani, Karl H. Johansson

Preprint, 2026

Offline policy improvement that explicitly constrains interaction-level distribution shift, evaluated in closed-loop driving and on a full-scale autonomous truck.

project pagearXiv
BibTeX 
@article{selim2026beyond,
  title = {Beyond Policy Support: Interaction-Constrained Offline Reinforcement Learning for Autonomous Driving},
  author = {Selim, Mahmoud and Cipriani, Cristina and Johansson, Karl H.},
  journal = {arXiv preprint arXiv:2610.09763},
  year = {2026}
}
NSPG illustrative placeholder

Noisy-Space Policy Gradient for Diffusion Policies in Offline Reinforcement Learning

Mahmoud Selim, Cristina Cipriani, Karl H. Johansson

ICML 2026

A principled policy gradient for diffusion policies that connects noisy latents to clean-action values without backpropagating through denoising.

project pagecode
BibTeX 
@inproceedings{selim2026noisyspace,
  title = {Noisy-Space Policy Gradient for Diffusion Policies in Offline Reinforcement Learning},
  author = {Mahmoud Selim and Cristina Cipriani and Karl Henrik Johansson},
  booktitle = {Forty-third International Conference on Machine Learning},
  year = {2026}
}
MetaKoopman illustrative placeholder

MetaKoopman: Bayesian Meta-Learning of Koopman Operators for Modeling Structured Dynamics under Distribution Shifts

Mahmoud Selim, Sriharsha Bhat, Karl H. Johansson

NeurIPS 2025

Bayesian meta-learning for fast, uncertainty-aware adaptation of dynamics models, with real-world validation on an autonomous truck in winter conditions.

project pagepapercode
BibTeX 
@inproceedings{selim2025metakoopman,
  title = {MetaKoopman: Bayesian Meta-Learning of Koopman Operators for Modeling Structured Dynamics under Distribution Shifts},
  author = {Selim, Mahmoud and Bhat, Sriharsha and Johansson, Karl H.},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {38},
  year = {2025}
}

Contact

I welcome conversations about machine learning, robot intelligence, and autonomous systems.

mahmoud.selim@scania.com