Benchmark RadarEBench
EBench is a simulation benchmark for diagnosing generalist mobile manipulation policies. It comprises 26 manipulation tasks annotated along five capability dimensions (scene, atomic skill, horizon, precision, mobility) and four generalization dimensions (object, background, instruction, mixed), with strict train/test splits and held-out online evaluation.
- Released
- 2026-06-16
- Readiness
- Runnable
- Primary field
- Robotics & Autonomous Systems
Why it matters
EBench addresses the need for multi-axis diagnostic evaluation of generalist manipulation models, moving beyond single success-rate metrics. It provides practical decision value by revealing capability and generalization profiles that guide model iteration and selection.
Motivation
We present EBench, a simulation benchmark that diagnoses generalist mobile manipulation policies beyond a single success-rate scalar.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.