Joy Ge

Challenge

Built a robot DJ that reaches small knobs, faders, and buttons precisely and executes transitions within a 4ms of latency.

Approach

Built the control repository from scratch, including teach.py for recording robot positions and logic for deriving new positions from Cartesian offsets. Baked in an inverse kinematic solver to convert target poses into valid joint configurations, then designed motion paths that retract safely, cross the controller quickly, and approach controls linearly for precise interaction.

Impact

Delivered a working robot DJ capable of transitioning between songs using EQ controls, hot cues, filters, buttons, and fader sliders. V2 will add software that detects what is playing and selects transitions dynamically based on the music.

Challenge

Redesign a robotic painting cell whose hardcoded components locations and 45-point calibration process made setup slow, difficult to reproduce, and vulnerable to probing error.

Approach

Owned the redesign from mathematical validation through physical deployment. Rebuilt the breadboard and every mounted component around self-locating mechanical features, including repeatable brush mounts, then assembled and commissioned the complete system. Developed a partial affine calibration algorithm that uses four probed board corners to solve three least-squares coordinate transforms, mapping CAD-defined board positions into each robot’s coordinate frame while correcting for its unique mounting offset. Replaced a mathematically weak centre-point check with two held-out, edge-biased verification points, an independent corner-geometry check, and pseudoinverse-based fault attribution to identify likely mis-probed corners. Integrated the calibration workflow into the Python robot controller and operator GUI, including incremental validation, pass/fail feedback, guided re-probing, and blended robot motion between calibration points. Used RoboDK simulations to verify reach, collision clearance, and optimal stroke lengths before deployment.

Impact

Reduced calibration from 45 measured locations to six probes while enforcing less than 2 mm worst-case validation error. Delivered a repeatable, CAD-driven production cell that operators could calibrate directly through the a GUI, eliminating manual measurement of component coordinates and improving the accuracy, serviceability, and reproducibility of the full robotic system.

Challenge

Build a robot arm and control it at a distance in real time.

Approach

Manufactured the robotic assembly and linkage-based EOAT with overextension capabilities for adaptive object picking. Programmed in Arduino/C++: motion-smoothing and dampening logic, debouncing sensor logic and controlling servo actuation with a PCA9685 driver for 6-axes of motion. Developed electrical schematics and performed circuit calculations for the robotic arm and wearable glove controller, including flex sensor integration, serial communication, and multi-axis motor control.

Impact

Achieved <1s delay multi-axis motion through custom Arduino control architecture, MPU6050/flex sensor input mapping, and capacitive filtering to reduce voltage fluctuations and servo jitter during operation.