VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models

Key points

• Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but incur high computational costs from processing long token sequences at every control step, limiting real-time deployment.1
• Visual token pruning offers a direct solution, as visual patches dominate the input sequence and contain considerable redundancy.1
• Experiments on LIBERO and real-world manipulation tasks show that VLA-ACL prunes up to 87.5% of visual tokens while retaining competitive performance, reduces computation by up to 75%, and achieves a 1.5x inference speedup.1
• These results establish a stronger performance-efficiency trade-off than existing frozen-VLA pruning methods and demonstrate the value of action-level supervision for visual token selection.1

Sources (1)

• [1]VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 10:47 AM

  We introduce VLA-ACL (Action Consistency Learning), which learns a lightweight visual token pruning policy through action-level supervision while keeping the base VLA model entirely frozen.

  Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but incur high computational costs from processing long token sequences at every control step, limiting real-time deployment.

Extractive summary: sentences quoted from the sources.