Student Projects

Teaching a robot to master a board game using Reinforcement Learning

Updated Project Description Our research project explores how physical robots can learn to play board games via Reinforcement Learning (RL). We currently employ OpenAI Gym environments with Stable Baselines3 to train an agent directly on our 2D gantry robot hardware integrated with ROS2. In parallel, we leverage NVIDIA’s Isaac Gym for a high-fidelity simulation environment, which allows us to iterate on policies before transferring them to the real robot. A key focus of the project is bridging the sim-to-real gap and reducing the time required for on-hardware training using sophisticated RL techniques. We explore multiple RL approaches—both in simulation and on physical hardware—to investigate robust policy learning strategies. RTX 4090 GPUs back these trainings. The project offers a unique opportunity to design, refine, and deploy RL algorithms in a real-world setting. We seek students to help improve our existing software and hardware pipelines by experimenting with various RL methods. The ultimate goal is to enable the robot to excel at board games against human opponents and other learned agents, showcasing advanced capabilities in real-world reinforcement learning. Show detailsaddremove