2 papers
cs.RO2026
VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models
Owen Du, Yang Yue, Jie Zhang +3
Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but incur high computational costs from processing long token sequences at every control step, l…
eess.SY2026
Policy-Level Recursive Self-Improvement for Embodied AI with a Criticality World Model
Linxuan He, Yuying Tian, Lingxiang Fan +5
Despite advances in policy pretraining, embodied AI systems can plateau during task-specific fine-tuning as uniform scenario collection encounters fewer of the remaining failures.…