Robot Harness for Coding Agents

A Simple Harness Outperforms VLA and World Action Models

With RoboHarness, Qwen3.8-Flash-Next outperforms state-of-the-art VLA models on BEHAVIOR Challenge 2025 tasks.

Workspaces as Robot Harness

Puts the robot harness into the coding agent's workspace: the agent freely adds, removes, edits and calls Python tools, reads observations, and can write its own scripts to control the joints.

Learn more Code

News

2026-10-25RoboHarness V2 open-sourced: Workspaces as Robot Harness.
2026-10-07Code open-sourced on GitHub.
2026-09-29Paper available on arXiv: RoboHarness: A Simple Harness Outperforms VLA and World Action Models.

Citation

@misc{qu2026roboharness,
  title         = {RoboHarness: A Simple Harness Outperforms VLA and World Action Models},
  author        = {Bince Qu and Wei Chen and Bo Zhang},
  year          = {2026},
  eprint        = {2610.14018},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  doi           = {10.48550/arXiv.2610.14018},
  url           = {https://arxiv.org/abs/2610.14018}
}