Balint Hodossy
Embodied AI Researcher, Imperial College London
I'm a postdoctoral researcher passionate about simulating motion. My work focuses on virtual environments, reinforcement learning for motor control and human-machine interfacing. My research interest is understanding human movement and building systems that support it.
I'd like to share what I've been up to on this site. If you'd prefer a summary, here's my CV.
Past Work
![If at first you don't succeed... [An image of simulated prosthesis user celebrating standing successfully while past versions surround them]](images/sim_pic_3.png)
PhD Thesis
Written after my 4 year PhD at Imperial. It is titled "Learning Steps: Models of Intent-Driven Lower Limb Prosthesis Use". It compiles the 4 manuscripts I wrote into one document, merging EMG-based intent estimation models with RL-trained locomotion models.
(It is shared under CC BY-NC 4.0.)
Shared Autonomy Locomotion Synthesis With a Virtual Powered Prosthetic Ankle
The core part of my project! I used RL to learn a gait policy for a humanoid agent to follow arbitrary walking paths. Then I introduced a second agent that controlled one of the ankles - representing an intent driven bionic limb. Much of the code of my RL environment construction tools are available open-source.
Full article in TNSRE Demo video Open-source project
Leveraging High-Density EMG to Investigate Bipolar Electrode Placement for Gait Prediction Models
A much needed address of the sensitivity of Deep Learning intent estimators to sensor location and configuration. I was surprised that simply sampling electrode placements randomly during training ended up working so well!
Full article in THMS
High-Level Locomotion Intent Estimation From Electromyography and Body Posture
Translating my biosignal processign models to high-level intent - so turns, stops, pivots instead of just straight walking. This abstract intent can be synthesized in simulation, which would mean not needing to simulate EMG for Sim2Real. Manuscript under review.
Read preprint
Towards a Virtual Gait Lab: Testing Prosthetics with Dynamically Simulated Subjects
I had the chance to work with Antonio Bicchi's robotics lab - together we modelled the IIT SoftFoot. I extended my locomotion agents to obstacles and slopes, and evaluated the advantage of the compliant foot. Manuscript under review.
Read preprintMy other publications can be found on Scholar. Also, many of my projects are documented and available at github.com/Balint-H.
Side projects
There are some side projects that I'm particularly fond of, or are not listed on my GitHub.
Simulating Wearable Robotics Workshop
This repo was the yearly evolving content for the workshop I hosted at the Summer School of Neurorehabilitation 2023-25. Now I use it when I need to introduce RL and physics simulation methods to someone. Usually I don't end up using all the content that gathered up all the years - there's examples for simulating a robotic hand, finding a controller for a powered ankle exoskeleton, implementing LQR controllers for balancing and much more!
See GitHub Repo
MyoChallenge 24/25
I was part of the organizing team of the MyoChallenge in 2024 and 2025. These were programming competitions at NeurIPS, tasking participants with controlling complex musculoskeletal models to achieve various tasks. Each year I contributed to designing and developing one of the challenges. In 2024 I helped with the bimanual manipulation task, and in 2025 I helped with the tabletennis task.
Visit website Preprint of MC24
Student Led Conference
During our first year of PhD students in my programme, we decided to organize a conference! Covid had different plans. We improvised, and managed to pivot to an online event. Turned out to be a great outlet for student research, and became a conference series.
Visit the debrief page
Roleplaying games
Each year I try to host 1-2 stories as a game master for my friends, using adventure plotlines I've written over the years. I'm working on releasing a module along with handouts such maps, newspapers, posters and puzzles I've made!
Video games
Occasionally, I practice coding by making video games. These range from multiplayer games made with python for my phone, to custom hardware projects made for the Arduino with potentiometers and microswitches as inputs. I have a couple prototypes that are available online.
Games on itch.io