FibVLA: An Efficient Temporal Vision-Language-Action Model with Fibonacci Sampling

2026-07-31

AI Summary

This work addresses the trade-off between long-horizon contextual modeling and real-time inference in vision-language-action models by introducing a logarithmic backtracking mechanism based on Fibonacci sampling. This approach efficiently fuses proprioceptive states with visual frames and leverages flow matching to generate action distributions. Furthermore, a Fibonacci recursive inference architecture is proposed, enabling closed-loop, feedback-driven long-horizon action planning without retraining large-scale visual encoders. The method significantly improves action smoothness and task success rates while demonstrating superior real-time responsiveness compared to video-based baseline models in real-world environments.

Abstract

Vision-language-action models (VLAs), which leverage the cognition of multimodal information to infer physical-world actions, provide a generalized solution for embodied AI applications. Conventional VLAs usually concentrate on current digital cognition. While some efforts are made to enhance VLAs' reasoning capabilities by capturing temporal information, encoding the long-context history causes an efficiency-decreasing issue. To reconcile the conflict between capturing temporal information and maintaining inference efficiency in VLAs, this paper introduces FibVLA, an efficient framework featuring temporal perception of long-context history. Specifically, we leverage logarithmic hindsight sampling to both proprioceptive states and visual frames to capture long-term temporal dependencies with minimal redundancy. For the action expert, we introduce the flow matching to produce action distributions, and the Fibonacci recurrent inference strategy to generate long-range planning steps based on real-time closed-loop feedback. Experiments demonstrate that FibVLA significantly improves action smoothness and success rates without retraining large-scale visual encoders. Efficiency analysis demonstrates superior real-time responsiveness compared to video-based baselines in real-world evaluations.

Problem

• vision-language-action models
• temporal information
• inference efficiency
• long-context history
• embodied AI

Innovation

• Fibonacci sampling
• temporal vision-language-action model
• logarithmic hindsight sampling
• flow matching
• recurrent inference