DEFT-VLA: Stage-Localized Force Specialization for Contact-Rich Robotic Manipulation
We propose DEFT-VLA (Dual-Expert Force-specialized Training for Vision-Language-Action Manipulation), which retains the task-adapted base expert for vision-dominated non-contact behavior and introduces a force-aware contact expert for critical contact stages, localizing force-conditioned adaptation to where it is needed.
ProofPaper ↗
Key points
- Contact-rich manipulation poses a stage-dependent sensory conflict: vision and proprioception are essential for localization and approach, when force is largely uninformative and potentially distracting, whereas force becomes critical after contact, when irrelevant visual details and correlated robot states may become shortcuts.
- We further introduce a contact-localized force-attending curriculum that applies force-attending post-training only to the contact expert, progressively restoring vision and masking state tokens to suppress shortcuts.
- Real-robot experiments on pipette tip attachment, bimanual pipette aspiration, and whiteboard wiping show that DEFT-VLA effectively exploits force feedback to improve contact-stage performance while maintaining strong non-contact behavior and robustness to held-out grasp heights.
- It improves overall task success by 20-40 percentage points over the strongest baseline on each task.
Sources (1)
- [1]DEFT-VLA: Stage-Localized Force Specialization for Contact-Rich Robotic ManipulationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 10, 12:14 PM
We propose DEFT-VLA (Dual-Expert Force-specialized Training for Vision-Language-Action Manipulation), which retains the task-adapted base expert for vision-dominated non-contact behavior and introduces a force-aware contact expert for critical contact stages, localizing force-conditioned adaptation to where it is needed.
Contact-rich manipulation poses a stage-dependent sensory conflict: vision and proprioception are essential for localization and approach, when force is largely uninformative and potentially distracting, whereas force becomes critical after contact, when irrelevant visual details and correlated robot states may become shortcuts.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 10, 2026PRISM: Precision Forcing for Generalist Vision-Language-Action Policies
- Oct 10, 2026Flow-Support-Weighted Proposals for Test-Time Verified VLA Policies
- Oct 10, 2026Red-Teaming of Vision-Language-Action Models via Quality Diversity Environment Generation
- Oct 9, 2026Memory Without Curation: Learning Multi-Stream Latent Memory for Robot Policies
- Oct 9, 2026Speech event detector: Timestamped detection of non-verbal vocalizations and background sounds for rich speech transcription
- Oct 8, 2026Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer