VABAR (Variational Automation Benchmark for Agentic Robotics) is the first open, simulation‑based test that asks robots to solve six manipulation challenges drawn from industry, medicine, and infrastructure – tool sorting, tool hanging, gear meshing, syringe packing, cable insertion, and cable routing. The benchmark defines each task by initial and goal configurations, an asset library, and robot specifications, then introduces systematic variations in objects, arrangements, and positions to create low‑ and high‑difficulty levels.

The implementation uses the Newton physics engine and the Isaac Lab Arena, giving researchers two interchangeable backends. A reusable library of 33 reusable skills and standardized success metrics lets teams plug in their own learning or planning pipelines. The framework stays simple enough that any lab can reproduce the exact conditions without spending months on simulation tuning.

Visual 1: Overview of VABAR’s modular task specification, variation space, and dual‑backend evaluation.