Thomas Power

I’m a Research Engineer at Google DeepMind.

I completed my PhD in Robotics at the University of Michigan where I worked at the ARM lab, advised by Dmitry Berenson. During my PhD, I worked on applying deep generative models (normalising flows, diffusion models) to accelerate planning & trajectory optimisation. I previously interned at Honda Research Institute USA where I worked on dexterous manipulation.

news

Jan 28, 2025	Our paper DIPS was accepted to ICRA 2025! DIPS combines diffusion models with search to generate contact sequences for contact-rich manipulation.
Nov 25, 2024	I joined Google DeepMind Robotics!
Jun 09, 2024	Our paper on optimizing diverse sets of constraint-satisfying trajectories was accepted to T-RO!
Jun 03, 2024	I successfully defended my Ph.D.!
Feb 05, 2024	Our paper on learning sampling distributions for sample-based MPC using Normalizing Flows was accepted to T-RO!
Dec 21, 2023	I am excited to be on the scientific committee for the Back to the Future: Robot Learning Going Probabilistic workshop at ICRA 2024
Oct 01, 2023	Our Differentiable Probabilistic Robotics workshop at IROS 2023 in Detroit was attended by over 200 registrants!

selected publications

Diffusion-Informed Probabilistic Contact Search for Multi-Finger Manipulation

*Abhinav Kumar , *Thomas Power , Fan Yang , and 4 more authors

In International Conference on Robotics and Automation (ICRA) , 2025

Bib HTML Video

@inproceedings{dips,
  author = {Kumar, *Abhinav and Power, *Thomas and Yang, Fan and Aguilera Marinovic, Sergio and Iba, Soshi and Soltani Zarrin, Rana and Berenson, Dmitry},
  booktitle = {International Conference on Robotics and Automation (ICRA)},
  title = {Diffusion-Informed Probabilistic Contact Search for Multi-Finger Manipulation},
  year = {2025},
  doi = {10.1109/ICRA55743.2025.11127844},
}

Learning a Generalizable Trajectory Sampling Distribution for Model Predictive Control

Thomas Power, and Dmitry Berenson

IEEE Transactions on Robotics, 2024

Bib HTML Video

@article{10449376,
  author = {Power, Thomas and Berenson, Dmitry},
  journal = {IEEE Transactions on Robotics},
  title = {Learning a Generalizable Trajectory Sampling Distribution for Model Predictive Control},
  year = {2024},
  volume = {40},
  pages = {2111-2127},
  doi = {10.1109/TRO.2024.3370026},
}

Constrained Stein Variational Trajectory Optimization

Thomas Power, and Dmitry Berenson

IEEE Transactions on Robotics, 2024

arXiv Bib HTML

@article{power2023constrained,
  title = {Constrained Stein Variational Trajectory Optimization},
  author = {Power, Thomas and Berenson, Dmitry},
  journal = {IEEE Transactions on Robotics},
  year = {2024},
  volume = {40},
  number = {},
  pages = {3602-3619},
  doi = {10.1109/TRO.2024.3428428}
}

Variational Inference MPC using Normalizing Flows and Out-of-Distribution Projection

Thomas Power, and Dmitry Berenson

In Robotics: Science and Systems , 2022

arXiv Bib Video

@inproceedings{power2022flowmpc,
  title = {Variational Inference MPC using Normalizing Flows and Out-of-Distribution Projection},
  author = {Power, Thomas and Berenson, Dmitry},
  booktitle = {Robotics: Science and Systems},
  year = {2022},
}

Keep It Simple: Data-Efficient Learning for Controlling Complex Systems With Simple Models

Thomas Power, and Dmitry Berenson

IEEE Robotics and Automation Letters, 2021

Bib HTML Video

@article{power2021lvspc,
  author = {Power, Thomas and Berenson, Dmitry},
  journal = {IEEE Robotics and Automation Letters},
  title = {Keep It Simple: Data-Efficient Learning for Controlling Complex Systems With Simple Models},
  year = {2021},
  volume = {6},
  number = {2},
  pages = {1184-1191},
}

Learning When to Trust a Dynamics Model for Planning in Reduced State Spaces

Dale McConachie , Thomas Power, Peter Mitrano , and 1 more author

IEEE Robotics and Automation Letters, 2020

Bib HTML Video

@article{8988145,
  author = {McConachie, Dale and Power, Thomas and Mitrano, Peter and Berenson, Dmitry},
  journal = {IEEE Robotics and Automation Letters},
  title = {Learning When to Trust a Dynamics Model for Planning in Reduced State Spaces},
  year = {2020},
  volume = {5},
  number = {2},
  pages = {3540-3547},
  doi = {10.1109/LRA.2020.2972858},
}