CompVLA: A Variable Compliance Vision-Language-Action Model for Contact-rich Manipulation
CompVLA: A Variable Compliance Vision-Language-Action Model for Contact-rich Manipulation
This research presents CompVLA, a unified framework that extends conventional Vision-Language-Action models to regulate both motion and compliance via stiffness matrix prediction. By integrating a dedicated Compliance Expert with geometric impedance control, the model outputs time-varying stiffness and virtual displacement profiles from RGB and language inputs. The authors report that CompVLA achieves the highest average success rate across diverse contact-rich tasks, outperforming vanilla and compliance-aware baselines, with ablation studies confirming the necessity of each component.
The paper introduces CompVLA, a Vision-Language-Action model that jointly predicts motion and stiffness matrices to handle contact-rich manipulation tasks, claiming superior success rates over baselines based on ablation studies.
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.
Dataset details reported by the source
- Modalities
- RGB and language inputs Open source S3
- Tasks
- contact-rich manipulation Open source S2
- Embodiments
- robots Open source S2
- paper: Reported pending Open source S2
- paper: not_reported
What remains unknown
Dataset licensing and commercial-use rights are not established here.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
jointly predicts motion and a stiffness matrix from RGB and language inputs. Our approach augments the conventional architecture with a dedicated Compliance Expert, which outputs time-varying stiffness and virtual displacement profiles executed via geometric impedance control. We demonstrate that CompVLA achieves the highest average success rate across diverse contact-rich tasks, outperforming both vanilla and compl
Open source S3
arXiv:2609.23614v2 Announce Type: replace Abstract: Contact-rich manipulation, requiring robots to regulate not only motion but also how they yield to external forces, has emerged as the next frontier for Vision-Language-Action (VLA) models. However, existing VLAs output purely kinematic commands, degrading performance on real-world contact-rich tasks. In this paper, we introduce CompVLA, a unified VLA framework that
Open source S2
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Check the exact dataset version, license, modalities and task coverage against the intended use.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：arXiv Robotics — research abstracts · arxiv.org