Focused on real world learning for hard to simulate manipulation problems.
RL-trained humanoid on inline skates fitted to a Booster T1. Custom Gym environment, realistic wheel dynamics, tuned reward shaping. Deployed sim-to-sim and sim-to-real, faster, more efficient, and more robust than prior humanoid-skater work, with no motion priors and no imitation learning.
Repurposed Verity's warehouse drone hardware and CV stack to track forklifts and register pallet moves. Field-deployed a prototype at a client facility, gathered reliability metrics, and built a simulated warehouse to stress-test failure modes. The work made the case for a hybrid drone-plus-forklift tracking system.
Designed a custom 11-DoF bio-mimetic hand: CAD, electronics, and digital twin model. Trained an RL policy in Isaac Gym to roll a ball about arbitrary axes at a target velocity. Deployed successfully on the physical hand, which could also be actuated via teleoperation.
A tilt-actuated marble maze with computer-vision path planning. Custom CAD and mechatronics for the two-axis tilt rig; a vision pipeline segmented the maze and computed a path for the marble to follow. Awarded NYU Tandon's 2023 Best Senior Design.
Mechanical and product-design consulting on IoT products at every life-cycle stage, across startups and large companies alike. Every project closed on time, and follow-on contracts drafted.