Mitanshu Goel
Robotics & AI Engineer
I turn human demonstrations into robot policies: teleop rigs, synchronised capture, and the π0.5 fine-tune. Three robotics internships, most recently Physical AI at Nferent AI. Open to Physical-AI and robotics roles.
See the work ↓ Read the CV →Real hardware, except the two clips marked simulation.
Three internships, a student lab, and my own time.
Two industrial arms copy my hands live, tracked by a Quest 3 hung on my chest. I wrote the real-time C++ loop: Cartesian servoing at 125 Hz, bad tracking frames dropped.
A 20-motor Tesollo hand. I built the vision side: a camera reads your gesture and the hand plays its move.
I teleoperated a Franka FR3 and an Elite CS66 to record a 10-task LeRobot dataset, then fine-tuned a π0.5 policy on it.
A Yanshee humanoid that spots objects with YOLOv8 and answers questions out loud, gated by a wake word.
I worked on motion planning for a six-joint robotic arm. I used ROS 2 and Gazebo for simulation, and MoveIt’s C++ interfaces to plan and test motions on the physical arm.
Two simulated robot dogs, one warned before each shove. At 250 N the warned one falls 5× less. Take its warning away and it falls 9× more than the one never warned.
A team walker from the A.T.O.M. lab at MAIT. Mine: the control node and the closed-form leg IK. It runs on the real robot; the clip is Gazebo.