README.md

metadata

license: mit
task_categories:
  - robotics
tags:
  - lerobot
  - vla
  - mobile-manipulation

EBench: Elemental Mobile Manipulation Benchmark

This repository contains the training trajectories (stored in LeRobot format) for EBench, an indoor VLA (Vision-Language-Action) manipulation benchmark built on NVIDIA Isaac Sim. Instead of compressing a model's behavior into a single overall success rate, EBench produces a multi-axis capability profile that exposes what a model is good at and where it overfits.

• Paper: EBench: Elemental Diagnosis of Generalist Mobile Manipulation Policies
• Project Page: EBench Homepage
• Documentation: EBench Docs
• GitHub Repository: InternRobotics/EBench

Key Features

• Three manipulation regimes in one benchmark — covers long-horizon, dexterous & precise, and mobile manipulation.
• 5-axis atomic diagnostic — every task is labelled by Scene · Atomic Skill · Horizon · Precision · Mobility.
• 4-axis generalization tests — controlled perturbations along Object · Background · Instruction · Mixed.
• LeRobot Format — trajectories are structured using Hugging Face's lerobot data format, making them ready for loading and training.

Citation

If you find EBench useful in your research, please cite:

@misc{ebench2026,
  title  = {EBench: Elemental Mobile Manipulation Benchmark},
  author = {Shanghai AI Laboratory},
  year   = {2026},
  note   = {Preprint coming soon},
  url    = {https://internrobotics.github.io/EBench-doc/}
}