Auriga Technical Report: A Task-Agnostic Agent Harness for Zero-Shot Embodied Manipulation
Auriga Technical Report: A Task-Agnostic Agent Harness for Zero-Shot Embodied Manipulation
This paper introduces Auriga, a task-agnostic agent harness designed to bridge the planning-execution gap in zero-shot embodied manipulation. Developed by Chengcheng Wang and colleagues, the system utilizes vision-language models for task decision-making while managing physical execution through calibrated 3D measurements and dual-arm control. The report highlights that Auriga closes 68.3% of the performance gap observed on RoboDojo without requiring target-task trials or resets. By inspecting visual trajectory previews and checking stage tolerances during execution, the harness achieves a 57.7% success rate on RoboDojo and a 73.2% success rate on RoboCasa365, outperforming reference harnesses and VLA-assisted agents in composite tasks.
The authors report a zero-shot embodied manipulation agent harness that closes a significant planning-execution gap on RoboDojo and RoboCasa365 benchmarks, with success rates attributed to the specific model and harness configuration described in the abstract.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
22.5%
View original evidence
arXiv:2611.01112v1 Announce Type: new Abstract: Vision-language models can interpret instructions and plan manipulation tasks, yet struggle when acting directly as robot policies. On RoboDojo, a large planning-execution gap separates high-level task decisions from physical execution: removing arm execution raises GPT-6 Astra's success rate from 22.5% under the reference control interface to 76.2%. We present Auriga,
Open source S2Reported success rate
76.2%
View original evidence
arXiv:2611.01112v1 Announce Type: new Abstract: Vision-language models can interpret instructions and plan manipulation tasks, yet struggle when acting directly as robot policies. On RoboDojo, a large planning-execution gap separates high-level task decisions from physical execution: removing arm execution raises GPT-6 Astra's success rate from 22.5% under the reference control interface to 76.2%. We present Auriga,
Open source S2Grasp-and-place success rate
57.7%
View original evidence
Before motion, the model inspects a visual trajectory preview while simulation is paused; during execution, the harness checks declared stage tolerances and grasp bounds, stopping the robot on divergence and returning the measured discrepancy for in-place recovery. Developed across 13 cross-task capability iterations and frozen for evaluation, Auriga ranks first on the RoboDojo leaderboard with a 57.7% success rate (
Open source S4Reported success rate
73.2%
View original evidence
64.0 Score), more than doubling the reference harness on the same model, and remains first under scene randomization while learned policies drop by up to 33.9%. On RoboCasa365, the same frozen harness reaches a 73.2% success rate, outperforming VLA-assisted agents by up to 23.1% on composite tasks. Project page: https://auriga.stelledge.com/ Code: https://github.com/lose4578/Auriga
Open source S5
- paper: not_reported
- code: Reported pending Open source S5
- code: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.01112v1 Announce Type: new Abstract: Vision-language models can interpret instructions and plan manipulation tasks, yet struggle when acting directly as robot policies. On RoboDojo, a large planning-execution gap separates high-level task decisions from physical execution: removing arm execution raises GPT-6 Astra's success rate from 22.5% under the reference control interface to 76.2%. We present Auriga,
Open source S2
a task-agnostic agent harness that closes 68.3% of this gap in zero-shot manipulation without target-task trials or resets. In Auriga, the model makes all task decisions over an explicit task belief and acts through either direct tool calls or synthesized multi-step Python programs, while the harness supplies calibrated 3D measurements, finger-referenced dual-arm control, and a single admission gateway to the robot.
Open source S3
Before motion, the model inspects a visual trajectory preview while simulation is paused; during execution, the harness checks declared stage tolerances and grasp bounds, stopping the robot on divergence and returning the measured discrepancy for in-place recovery. Developed across 13 cross-task capability iterations and frozen for evaluation, Auriga ranks first on the RoboDojo leaderboard with a 57.7% success rate (
Open source S4
64.0 Score), more than doubling the reference harness on the same model, and remains first under scene randomization while learned policies drop by up to 33.9%. On RoboCasa365, the same frozen harness reaches a 73.2% success rate, outperforming VLA-assisted agents by up to 23.1% on composite tasks. Project page: https://auriga.stelledge.com/ Code: https://github.com/lose4578/Auriga
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org