Robotics

Real-world reinforcement learning, sim-to-real transfer, and accessible robotics platforms

Projects & Highlights

BrickSim — From Bricks to MuJoCo and Back

BrickSim — LeoCAD to MuJoCo

CAD models of BricksRL robots designed in LeoCAD, converted to MuJoCo simulation models, and trained in parallel on GPU with MuJoCo MJX for sim-to-real policy transfer.

Video 1 Video 2
BricksRL RoboArm

BricksRL

A platform for democratizing robotics and reinforcement learning research and education with LEGO. Build real robotic systems with accessible hardware and train RL agents on physical robots.

Paper Project Page GitHub

Offline RL for RoboArm

Offline reinforcement learning on BricksRL roboarm tasks with image observations, trained on small real-world datasets without further environment interaction.

Video Short 1 Short 2

Autonomous RC Car

A CNN-controlled RC car that learns to drive autonomously on a race track using end-to-end convolutional neural network inference.

Video GitHub

Happy to discuss any of these — feel free to reach out.

Videos & Media

Demos, experiments, and robot builds

Gallery

Photos and additional media

LeoCAD 2Wheeler model
2Wheeler robot modeled in LeoCAD
2Wheeler in MuJoCo
2Wheeler imported into MuJoCo simulator
Robot view — green objects
Real-world observation: green object detection
Robot view — yellow objects
Real-world observation: yellow object detection