ROBOGATE: Adaptive Failure Discovery for Safe Robot Policy Deployment via Two-Stage Boundary-Focused Sampling
RoboGate maps simulated robot-policy failure boundaries and benchmarks a scripted pick-and-place controller across four arm embodiments.
Why it matters
Deploying learned robot manipulation policies in industrial settings requires rigorous pre-deployment validation, yet exhaustive testing across high-dimensional parameter spaces is intractable. We present ROBOGATE, a deployment risk management framework that combines physics-based simulation with a two-stage adaptive sampling strategy to efficiently discover failure boundaries in the operational parameter space.
Evidence notes
- The UR3e configuration uses a suction SurfaceGripper and 10,000 Latin-hypercube-sampled Isaac Sim experiments.
- UR3e achieved 9.6% success in the shared workspace; the study attributes the low rate primarily to its 0.50-metre reach placing tasks at the reachable boundary.
- All UR3e failures were grasp misses; the reported evaluation is simulation-only and does not establish physical deployment performance.
Company context
Universal Robots develops collaborative robot arms for flexible automation in industrial and commercial environments.