Event
MAST-VLA Multi-Skill Manipulation Framework Paper Released
1 reports · 1 sources
Jiachi Yang and other authors submitted a preprint proposing the MAST-VLA framework for multi-skill robotic manipulation: it introduces skill-conditioned LoRA mixture-of-experts adapters into an LLM backbone, and converts predicted affordance heatmaps into a patch-level trajectory affordance spatial prior for the action head. The authors report that under the CALVIN ABC→D long-horizon multi-skill cross-scenario setting, the framework achieves an average completed length of 4.65 on five-subtask sequences, attaining the best performance in this setting; real-robot multi-skill sequence experiments further validate its deployability under real perception and control noise. The paper has been accepted to CoRL 2026.
Report timeline
- MAST-VLA: Multi-Skill Robot Manipulation with Skill-Conditioned Experts and Trajectory-Affordance Priors
MAST-VLA proposes a dual-conditioning framework that introduces skill-conditioned LoRA Mixture-of-Experts adapters into the LLM backbone and converts predicted affordance heatmaps into a patch-level trajectory-affordance spatial prior for the action head. In the CALVIN ABC→D long-horizon multi-skill cross-scene setting, it achieves an average completion length of 4.65 over five-subtask sequences, the best performance in that setting. Real-robot multi-skill sequence experiments further validate its deployability under real perception and control noise. The paper has been accepted to CoRL 2026.
arXiv cs.RO ·