Robotics Ph.D. candidate at UT Dallas · I try to make robot arms do useful things outside of simulation.
I work in the Intelligent Robotics and Vision Lab with Prof. Yu Xiang, on manipulation that has to survive contact with the real world — motion planning and trajectory optimization, learning tasks from human video, and the unglamorous work of figuring out why a pipeline that looked fine in sim drops the mug on hardware.
Before this: M.S. in Control & Automation at IIT Delhi, and co-founder at VECROS, where we flew drones through GPS-denied spaces and learned that reactive planners fail in far more creative ways than papers suggest.
One human demo video in, a mobile manipulation task out. VLM-grounded object reasoning plus trajectory optimization, no teleop and no per-task data collection.Best Poster, ECS AI Days 2026 · under review |
A benchmark for real-robot manipulation you can actually reproduce — fixed scenes, fixed object set, fixed protocol. 2,000+ trials comparing perception, grasping, and control stacks head to head.ICRA 2024 (Oral) |
Picking the grasp and planning the arm motion as one optimization problem over point clouds, instead of two separate ones that disagree. ~66% faster than an OMPL sampling-based baseline on a Fetch.IROS 2024 (Oral) |
A robot explores an unseen space on its own, builds a semantic map with open-vocabulary detection, and keeps it current as objects move around.under review |
Also: CasADi, OMPL, Gazebo. Hardware I've broken and fixed: Fetch, Franka, custom quadcopters, a 12-camera OptiTrack rig.
Open to relocation · Richardson, TX