Open-source framework to evaluate AI models on real robots
github.com/robocurve/inspect-robots
Inspect Robots is Robocurve's open-source evaluation framework for physical AI. It runs LLM/VLA and related policies on robot arms and humanoids against real or simulated benchmarks, with adapters for embodiments such as Franka, WidowX, Unitree G1, and AgiBot A2. Released under an open-source license on GitHub.