Isaac Rudnick

Undergraduate thesis. Tests whether introducing a simulated 4-DoF arm's sensors in a developmental order, as animals' senses come online, speeds up reinforcement learning. Against all-sensor and random-dropout baselines across three task phases (PPO, PyBullet), the ordered curriculum trained about 4x faster early on with matched final success.

A serial communication and control framework bridging Python and Arduino for robotics research: PWM servos, stepper drivers, and sensors through modular classes, so new components need minimal code on either end. Its showcase application, VectorPlotter, draws SVG images on a whiteboard; Vassar's robotics lab forked both and installed a plotter.

Six simulations for a 16×16 WS2812B LED matrix on an ESP32 (Game of Life, Langton's Ant, Brian's Brain, fire, rain, sleep) on a custom KiCad PCB with five buttons and a 3D-printed enclosure; state persists in EEPROM. An alternative fork has an RTC-powered sloth tamagotchi.

Shows real-time blood glucose from the Nightscout API on a 32×8 WS2812B matrix driven by an ESP32. Custom 3D-printed enclosure with a diffuser grid, brightness profiles, and day/night modes; an open-source alternative to commercial displays.

A browser-based experiment testing whether losing control narrows attention and whether high stakes make it worse, using a Frogger-style task with interruptions. Designed and run with Jackson Chidiac, Irene Kim, and Beatrice Messaris at Vassar, spring 2025.

Screens SEC Form 4 filings for clusters of open-market insider purchases and scores them, with a backtester for strategies built on those events. A research triage tool, not a trading system; the backtests document where the signal doesn't survive trading costs.