GabAI

What it does

GabAI records a user-guided motion on its 3-DOF arm, extracts the gripper’s key poses, refines the route with a genetic algorithm, and replays the job at ≥95 % repeatability—cutting programming time, strain injuries, and integration cost for small producers.

Your inspiration

Repetitive manual labor at our partner SME exposed staff to overexertion and contact-injury risks which are common in 1M+ U.S. cases yearly. Traditional robots that could help are too expensive and rigid for SMEs. Observing this gap, the team aligned with UN SDG 8’s call for decent, productive work and envisioned a cobot that anyone could “teach by touch.” Interviews with the SME owner confirmed the need for quick setup, safe human-robot sharing, and high repeatability. These insights—plus class research on kinaesthetic teaching limits—sparked the idea of pairing physical guidance with metaheuristic path optimization.

How it works

A user physically steers the 3-DOF arm (base, shoulder, elbow) while three 10 kΩ potentiometers sample joint angles at 20 Hz. The Raspberry Pi relays data to a Next.js/React PWA, which isolates gripper-angle events as task markers, subsamples points with a Gaussian-mixture filter, then feeds them to a genetic algorithm that iteratively breeds shorter, smoother trajectories. The chosen path is streamed back to an Arduino-CNC shield driving three NEMA 17 steppers and a gripper servo; a software e-stop and hardware cutoff meet ISO 13850, while force limits follow ISO/TS 15066 for safe collaboration. A status-LED array and buzzer signal power, WebSocket link, recording mode, and motor activity, and the PWA plots raw vs. optimized paths so operators can verify results before pressing “Run.”

Design process

Ask & Research: Interviews and task observations at the partner SME framed the need to automate repetitive placement while preserving artisan quality. Constraints on efficiency, energy, storage, maintainability, and risk were set. Imagine: Three alternative solutions—Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Biogeography-based Optimization (BBO)—were brainstormed and benchmarked against those constraints. Plan: A Pareto trade-off analysis selected the GA variant as most resilient considering multiple realistic-constraints. Create: Hardware prototypes combined 3D-printed links, NEMA 17 motors, an Arduino UNO +CNC shield, and a Raspberry Pi inside a ventilated “GabAI Box,” while the PWA, genetic optimizer, and safety interlocks were coded and integrated. Test: ≥20-run trials measured repeatability and stabilization time; the client joined hands-on sessions to vet usability and accuracy. Improve: Feedback prompted firmware tuning, UI refinements, and enclosure airflow tweaks until the system met >95 % task success and ISO safety metrics.

How it is different

Unlike industrial arms that require code-level reprogramming, GabAI lets non-experts “show” a task once, then automatically optimizes it—combining kinaesthetic learning, multi-algorithm metaheuristic search, and a browser-based controller in one low-cost kit. Prior platforms do some of these pieces, but none merge open-source Raspberry Pi/Arduino hardware, cloud-ready PWA supervision, and genetic-driven path refinement tailored to SME budgets. The design’s modular boards mean off-the-shelf repairs and easy DOF upgrades, while its visual path-verification plot builds user trust that black-box optimizers usually lack. Safety is baked in via dual e-stops and ISO/TS 15066 force limits, features often optional on entry-level arms.

Future plans

Next, the team will add extra joints and vision-ready toolheads so GabAI can tackle welding, painting, and pick-and-place variants without rewriting the optimizer. Continuous firmware and PWA updates will keep the system aligned with evolving safety standards and improve energy-aware trajectory tuning. On the business side, the team will package hardware kits for schools, offer cloud subscriptions for path analytics, and demo at expos with hardware partners to seed adoption in education and light-industrial niches.

Awards