Mitanshu Goel

B.Tech, Electronics & Communication Engineering · Minor in AI & Machine Learning · CGPA 8.01 / 10

Gurugram, India

• Fine-tuned a π0.5 vision-language-action policy on a 10-task LeRobot dataset collected across two arms, and worked on its inference path.
• Wrote the real-time C++ teleop loop that collected it, on Elite CS66 and Franka Research 3 arms: Cartesian servoing at 125 Hz, Quest 3 hand pose over UDP, bad tracking frames dropped.
• Built a capture rig keeping two MANUS gloves and three RealSense cameras on one timebase: drift under 15 ms at p95, and a watchdog that fails an episode if a camera repeats frames.
• Tesollo DG-5F hand (20 motors): built the vision layer, a MediaPipe finger-extension classifier that plays rock-paper-scissors, and root-caused a segfault in the vendor SDK's connect path.

Delhi, India

• Turned a UBTech Yanshee humanoid into a vision assistant with a custom-trained YOLOv8 detector on its MJPEG stream.
• Wrote the voice side: NVIDIA NeMo ASR routing the wake word and commands to a chat service or QR scanner.
• Built a sensor workstation for food supply-chain telemetry: sensor network, firmware and collection code, streaming over WiFi.
• Built a camera-driven pick-and-sort line on an arm and conveyor, sorting by colour, shape and class.

Ghaziabad, India

• Stood up a supplied 6-DOF arm in ROS 2: URDF links, joints and collision geometry, checked in Gazebo and RViz.
• Configured MoveIt with KDL IK for Cartesian and waypoint paths, validated in simulation, then fixed the URDF mismatches breaking trajectories on the real arm.

ROS 2 Humble · ros2_control · Gazebo · Docker

• Wrote the control node: a /cmd_vel command becomes a tripod gait through ros2_control, in Gazebo and on the real robot.
• Wrote the closed-form (law-of-cosines) leg IK, foot target to coxa, femur and tibia angles, replacing a hand-tuned angle table.
• Containerised the ROS 2 Humble and Gazebo Harmonic stack in Docker, with GPU and X11 passthrough, so the team shared one environment.

MuJoCo · mjlab · PPO (rsl_rl) · Unitree Go1

• Built the task on mjlab: a random shove every few seconds, plus a warning of its direction, force and time to impact. Difficulty rises per robot as it learns.
• Trained two PPO policies, 4096 parallel envs and 393M steps each, differing only in whether the actor sees the warning.
• With the warning removed at test time, the warned policy falls 52.3% of the time instead of 1.2%: it had learned to depend on it.

Student robotics society

• Build the team's robots and represent MAIT at competitions. Built a browser page that drives a robot arm over a rosbridge WebSocket, with its camera feed live.