opc

RealityTest is an open-source visual testing platform for real-world physical tasks. Robotics developers can test software in simulation, but verifying what actually happened in a physical environment often requires manual video review or custom instrumentation. RealityTest turns a short video into an explainable test result. Users upload a recording, identify the object to track, define a target region, and specify an expected outcome. Meta SAM 3.1 segments and tracks the object across video frames. A deterministic evaluation engine then analyzes its movement relative to the target region and returns PASS, FAIL, or INCONCLUSIVE. Instead of producing an opaque AI judgment, RealityTest provides visual evidence, object trajectories, and the frames supporting each result. The initial use case is pick-and-place task verification. The architecture is designed to support broader physical workflow testing, including robotics development, manufacturing demonstrations, and quality assurance. RealityTest is open-source, with a proposed paid cloud service for teams that want hosted video processing and evaluation history.