Self-Driving Labs for Soft Robotics

Motivation & problem

Developing new soft robotic actuators is slow because fabrication, testing, and characterization are largely manual: a researcher casts an actuator, waits for it to cure, mounts it by hand, runs pneumatic tests, and manually estimates curvature from video or photos before deciding on the next design iteration. That loop limits how much of the design space researchers can actually explore.

My contribution

As a software researcher on this project at the Nemitz Lab, I'm building the software system — EvoFab — that closes this loop. My contribution spans the full stack: a PostgreSQL/Supabase schema for tracking experiments and design iterations, a FastAPI backend that coordinates fabrication and testing steps, integration with a UR7e robot arm and Robotiq Hand-E gripper for automated fabrication handoff, and a computer vision pipeline (OpenCV, RGB-D camera) for shadow-invariant actuator segmentation and curvature regression — replacing manual measurement with automated characterization.

Current status

The system is actively being developed and used in the lab for real fabrication and testing cycles. A paper describing the system and its results is in progress and has not yet been submitted or published.

Abstract & PDF