RoboHarness lets an LLM agent beat VLA models on robot tasks
Original title: RoboHarness: A Simple Harness Outperforms VLA and World Action Models
The open-source harness gives agents a visual-geometric control panel; Qwen3.8-Flash-Next leads BEHAVIOR Challenge 2025.
RoboHarness is a robot control harness that lets a coding LLM agent see geometry and invoke embodied tasks instead of training a vision-language-action model. The authors argue a simple harness can support generalist manipulation without an action head trained on robot data. They report that Qwen3.8-Flash-Next with RoboHarness outperforms state-of-the-art VLA models on BEHAVIOR Challenge 2025 tasks. The implementation is open source.
Unconfirmed: Outperformance is the authors' claim against unnamed state-of-the-art VLA models.
- The harness is a visual-geometric control panel for LLM agents.
- Authors contrast it with scaled VLA and world action models.
- Evaluation is on BEHAVIOR Challenge 2025 tasks.
- The implementation is released as open source.
- Products
- RoboHarness
- AI models
- Qwen3.8-Flash-Next
From the original article
1 min read at huggingface.co
The rise of large language models (LLMs) has opened a path toward generalist embodied AI, raising the question of how LLMs can act in the physical world. One straightforward approach is training the model to generate actions: collect robot-action data and video demonstration, refine LLM architecture with an action head, and train a vision-language-action (VLA) model at scale.
By Bince Qu, Wei Chen, Bo Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.