Projects

python JAX GNN robotics deep-learning

Tensegrity GNN Simulator

Graph Neural Network physics simulator for NASA's tensegrity robot, built with JAX and Jraph. Learns complex cable-rod dynamics to enable fast rollouts for RL and model-predictive control.

GitHub Paper

python reinforcement-learning pytorch mujoco

Policy Gradients & MuJoCo

Vanilla Actor-Critic policy gradient algorithm on a custom MDP. Small neural networks for policy and value function, trained in a MuJoCo physics environment with Gaussian action noise.

GitHub

python robotics motion-planning mujoco

Kinodynamic RRT

Motion planner for systems with dynamic constraints. Implements Kinodynamic RRT in MuJoCo — grows a tree in state-control space using forward simulation, achieving 100% success rate at 20s time budget.

GitHub