VersatIL
Author · Open source
A modular PyTorch framework for imitation learning. VersatIL separates data processing, algorithms, policy architectures, and loss functions so that researchers can compose and compare policies across robots and sensing modalities. It supports standardized training and inference, with testing and continuous integration.
GitHub repository · Documentation
Sensorium
Lead developer · In development
A Python framework with a browser-based interface for recording, monitoring, synchronizing, annotating, and processing multimodal sensor data into episodic robot-learning datasets. Supports ROS 1 and ROS 2.
Paper accepted at GenAI4Health, NeurIPS 2026.
Edge inference for surgical robot learning
Physical demonstrator · In development
A demonstrator for gAIn – Next Generation AI Computing, investigating execution of surgical imitation-learning and Vision-Language-Action policies on resource-constrained edge hardware. Target platforms include Raspberry Pi 5 and Axelera METIS AIPU, using quantization, pruning, and optimized inference.
Open-H Embodiment dataset
Contributor
I contributed to the data collection pipeline and acquisition of 200 in vivo surgical trajectories for the NVIDIA Open-H Embodiment community healthcare robotics dataset.
Autonomous laparoscopic research platform
Co-main author · IROS 2026, accepted
An open-source Remote-Center-of-Motion controller and full-stack ROS platform connecting robot control, demonstration recording, and learned policy deployment.