[Submitted on 16 Nov 2026 (v1), last revised 17 Nov 2026 (this version, v2)]
Authors:Qixuan Wu (1), Xiaole Yang (2), Yuguang Yang (3), Canyu Chen (4), Linlin Yang (5), Haixia Pan (2), Guodong Guo (6), Baochang Zhang (1) ((1) School of Artificial Intelligence, Beihang University, Beijing, China, (2) School of Software, Beihang University, Beijing, China, (3) School of Electronic and Information Engineering, Beihang University, Beijing, China, (4) National College for Excellent Engineers, Beihang University, Beijing, China, (5) Communication University of China, Beijing, China, (6) Lane Department of Computer Science and Electrical Engineering, West Virginia University, Morgantown, USA)
Abstract:Pretrained video representations provide rich temporal information for robotic manipulation, yet effectively integrating them into existing vision-language-action (VLA) policies remains challenging. We present Flow-Aware Visual Residual Alignment (FAVR), a lightweight framework for flow-aware visual conditioning of frozen VLA policies. FAVR maintains a compact temporal visual state through linear Test-Time Training (TTT) and retrieves action-conditioned visual features throughout flow-based action generation. By exploiting the affine structure of the frozen action head, we explicitly characterize how injected visual features influence predicted action velocities and formulate flow-aware channel selection as a box-constrained optimization problem. We develop a FISTA-based solver for joint gate optimization and derive an efficient closed-form approximation through diagonalization and rescaling. With only 1.74-2.19M trainable parameters and 5,000 adaptation steps, FAVR improves pretrained VLA policies across benchmarks. On RoboTwin 2.0, FAVR improves LingBot-VLA from 87.12% to 90.04% under clean conditions and from 86.88% to 89.38% under randomized conditions. A controlled ablation on five RoboTwin tasks further shows that flow-aware gating increases average success from 72.0% with a gate shared across flow steps to 82.0%, demonstrating the benefit of adapting visual conditioning throughout action generation.
Submission history
From: Yuguang Yang [view email]
[v1] Mon, 16 Nov 2026 12:09:06 UTC (5,695 KB)
[v2] Tue, 17 Nov 2026 11:47:08 UTC (5,827 KB)
