Democratizing Robot-Use Agents: A Unified Harness with Open Models
Abstract
Uni-robot-agent enables open-weight robot-use agents to surpass proprietary models in zero-shot manipulation and suggests joint development of four core capabilities for future training.
Robot-use Agents (RUAs) have shown promising zero-shot manipulation capabilities, but reliance on frontier, proprietary models limits research into further training. To democratize RUA research, we introduce uni-robot-agent, a unified harness for competitive open-weight robot use that supports further training. Without privileged inputs, an out-of-the-box Qwen3.8-Flash-Next achieves a RoboDojo Score of 31.39, surpassing the published GPT-6 Astra-RoboProbe result of 28.97 and many VLA/WAM baselines, and performs competitively on 19 real-world tasks. Our analysis across model generations, families, and sizes links effective robot use to four general capabilities, suggesting that future RUA training should develop them jointly.
Community
We introduce uni-robot-agent, a unified harness for robot-use agents with open-weight models, designed to support further training.
Without privileged inputs, out-of-the-box Qwen3.8-Flash-Next achieves a RoboDojo Score of 31.39, exceeding the published GPT-6 Astra–RoboProbe result of 28.97, while also performing competitively on 19 real-world tasks.
Beyond benchmarking, we systematically analyze models across generations, families, and sizes, identifying four general capabilities linked to effective robot use. Our findings suggest that future robot-use agent training should develop these capabilities jointly.
Thank you for sharing your work on a unified harness for robot-use agents. Your unified harness enables an open-weight Qwen3.8-Flash-Next model to perform robot use without privileged inputs, reporting a RoboDojo score of 31.39 and competitive results on nineteen real tasks, with analysis of general capabilities, depth, and skill access.
We would like to discuss your work in Section 3.4 (“Skill- and agent-based execution”) of our survey, In-Context Learning for Robots: Methods and Applications, to discuss how general-model competence is connected to robot execution through tools and feedback, using the skill-library control to identify the contribution of stored routines.
If you find the discussion of skill-based execution helpful in situating your work, we would be grateful if you would consider citing the survey.
Get this paper in your agent:
hf papers read 2611.01560 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper