Mitanshu Goel
Robotics & AI Engineer
I build the pipelines that turn human demonstrations into robot policies — VR teleop rigs, multi-camera capture, and the datasets they produce. Physical AI at Nferent AI; before that SarthakAI and NextUp Robotics. Open to Physical-AI and robotics roles.
See the work ↓Every clip is real hardware. Press play. Filter by where it was built.
Where each project was built — three internships and my own time.
Two industrial arms copying my hands live through a Meta Quest 3, built at Nferent AI on a real-time C++ control stack I wrote — with a clutch so I can re-grip mid-task without the robots wandering.
A five-finger, 20-motor hand I programmed at Nferent AI — it reads your gesture through a camera and throws its own move back.
Driving a Franka research arm at Nferent AI behind a safety stack I wrote, recording a 51-episode manipulation dataset — 2.1 hours of RGB-D — in LeRobot format.
A small humanoid I built at SarthakAI — it spots objects with YOLOv8 and answers questions by voice, gated by a wake word.
My first internship, at NextUp Robotics — I wrote the MoveIt interfaces (joint, Cartesian-pose, and waypoint motion) with KDL inverse kinematics for a 6-DOF arm, checked in simulation, then run on the real hardware.
I told one simulated robot dog that a shove was coming and left an identical one in the dark, then shoved both about 148,000 times. The warned one falls 5× less often at 250 N — and 9× more often than normal once the warning is taken away.
An 18-DoF hexapod walker — the inverse-kinematics gait engine that turns a foot target into joint angles is mine. Walking here in Gazebo simulation.