projects

Simulation, control, and learning projects in robot manipulation — impedance control, motion planning, and reinforcement learning.

research

VR Teleop Data Collection for Franka Panda

Meta Quest 3 Pro teleoperation for a real Franka Panda, built to collect the demonstrations that train downstream manipulation policies

Latent Safety Filters: A Stability-Centric Safety Monitor for Manipulation

A label-free safety monitor that flags unsafe actions from a learned manipulation policy before they execute, by checking the stability of a learned world model's rollouts

Diffusion Policy for Real-Robot Manipulation

Adapting Diffusion Policy to dual-camera Franka Panda manipulation, with a real-time serving architecture for closed-loop control

Cartesian Impedance Control for Franka Panda

Null-space projected Cartesian impedance controller for a 7-DOF Franka Panda, simulated in MuJoCo with Pinocchio for kinematics/dynamics

LQR-Trees for Feedback Motion Planning

Lyapunov funnels and sums-of-squares programming to build regions of attraction and stabilize trajectories on a sparse randomized tree

Chaos-Aware Sensitivity Analysis of RL Policies

Using Finite-Time Lyapunov Exponents to analyze the sensitivity of SAC/PPO policies trained on a sparse-reward pendulum swing-up task