Hardware & Embedded Systems
Physical builds that run on real hardware
The robot balances itself on two wheels, localizes itself, and navigates a 2D environment autonomously. Achieved using full-state feedback LQR control, dead-reckoning for localization, and a wall-following FSM, all written in embedded C on a TI C2000 microcontroller.
A 3D-printed hypocycloidal gearbox paired with a brushless DC motor, delivering ~20 Nm of output torque for roughly $50, with a hall-effect sensor read over I²C for ~0.5° output-shaft accuracy.
A 3-DOF robot arm simulator on an FPGA: SystemVerilog hardware (a custom CORDIC trig unit, isometric projection, and an HDMI renderer) driven over AXI by inverse-kinematics and pick-and-place game logic in C on a MicroBlaze soft core.
Streams hand-tracking from Snap Spectacles AR glasses over a relay server to a physical LeRobot SO-101 arm: a pinch drives the gripper live, and a DH-model inverse-kinematics solver turns x/y/z targets into servo commands for the arm's joints. Built at CalHacks 12.
Research
Human-in-the-loop control
A custom instrumented seat and backrest let a person steer a simulated mobile robot through a VR environment using only their body's motion — validated against motion-capture and force-plate ground truth. Presented as a poster at the UIUC Undergraduate Research Symposium.
A ROS2 + Gazebo simulation where live BLE IMU torso motion is mapped directly to a robot's joints, with collision detection, sector timing, and dynamic obstacles in the course.
Autonomy & Controls
Planning and control, in simulation
A three-layer navigation stack for a planar differential-drive robot: A* path planning, Timed Elastic Band trajectory optimization with receding-horizon obstacle avoidance, and cubic-spline tracking with a modified PI controller — integrated with the team's existing simulator and backed by 151 passing tests. Team project (7 people).
A 2D drone path-following study on a spline-generated racetrack, quantitatively comparing an LQR-plus-artificial-potential-fields controller against a constrained MPC controller on tracking error, control effort, and safety-constraint satisfaction — backed by formal controllability and stability proofs.