CoTinyVLA: Chain-of-Thought Distillation for a Sub-Billion-Parameter Vision-Language-Action Model

2026-07-28

AI Summary

This work addresses the challenges of deploying large-scale vision–language–action (VLA) models on embedded robotic systems, where excessive parameter counts and insufficient robustness under perturbations hinder practical application. The authors propose a lightweight yet highly robust VLA framework with only 0.9 billion parameters, built upon the Qwen3.5-0.8B backbone. Key innovations include dual-view temporal input spanning 16 frames, instruction paraphrasing augmentation (scaling from 40 to 800 demonstrations), and a novel hierarchical chain-of-thought distillation mechanism that decouples knowledge from a 35B teacher model into task-level planning (Plan) and sub-task reasoning (Think). Evaluated on the LIBERO-Plus benchmark, the model achieves success rates of 90.8%, 87.3%, 86.6%, and 80.7% across four tasks—surpassing the strongest 7B baseline—and attains 73.6% on the most challenging initial-state task, all while requiring only 2.25 GiB peak GPU memory during inference.

Abstract

Vision-Language-Action (VLA) models translate natural-language commands into robot action sequences, but leading systems on the LIBERO-Plus robustness benchmark use three- to seven-billion-parameter backbones whose memory demands can exceed embedded robotic budgets. We present CoTinyVLA, a 0.9B-parameter action model on a Qwen3.5-0.8B backbone that obtains that robustness by structuring supervision instead of enlarging the model. Three components target different axes of the problem: dual-view temporal input of 16 history frames per step with textual camera and time markers; hierarchical chain-of-thought (CoT) distillation from a 35B teacher into an episode-level Plan and a chunk-level Think span over task phase, gripper state and next subaction; and paraphrase augmentation expanding 40 base commands into 800 variants. On LIBERO-Plus, spanning 10,030 perturbed tasks across seven perturbation dimensions, CoTinyVLA reaches 90.8% on Spatial, 87.3% on Object, 86.6% on Goal and 80.7% on Long, leading the strongest 7B baseline on all four suites by 4.7, 2.8, 15.9 and 3.0 points, with every margin interval excluding zero. The gains concentrate on the hardest axes of the benchmark: across the eleven published baselines none exceeds 53.2% on Robot Initial States in any suite, whereas CoTinyVLA reaches 73.6% on Goal against 39.9% for the strongest baseline. Ablations show the three components to be separable by perturbation axis, and at a matched image budget how frames are divided between the two cameras and across time accounts for 8.6 points on its own. Closed-loop inference peaks at 2.25 GiB of allocated GPU memory, and paired interventions show the episode Plan to be load-bearing: replacing it with an empty or contradictory span costs 40 to 45 points of success. Structured supervision thus lets a 0.9B backbone exceed all of them. Code: https://github.com/BrainJellyPie/CoTinyVLA

Problem

• Vision-Language-Action
• model compression
• robotic memory budget
• robustness benchmark
• parameter-efficient learning

Innovation

• Chain-of-Thought Distillation
• Vision-Language-Action Model
• Structured Supervision
• Temporal Dual-View Input
• Paraphrase Augmentation