Closed-Loop Neural Activation Control in Vision-Language-Action Models
Quick Take
CTRL-STEER introduces a closed-loop control framework for Vision-Language-Action models, enhancing stability and task success without retraining. It outperforms fixed-coefficient methods in regulating concepts during dynamic tasks, as demonstrated on four LIBERO task suites.
Key Points
- CTRL-STEER replaces static intervention with adaptive, time-varying control signals.
- Decouples representation from regulation for improved task performance.
- Utilizes both PID and reinforcement learning controllers for implementation.
- Achieves better steering-task success trade-off than fixed-coefficient baselines.
- Demonstrated on a fine-tuned OpenVLA policy across four LIBERO task suites.
Article Content
From source RSS / original summaryarXiv:2606. 00269v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models can be steered at test time by intervening on semantically meaningful internal directions, but existing methods use a fixed steering coefficient, effectively operating in open loop. This is poorly suited to embodied control, where task state and concept error evolve over time, often causing overcorrection, oscillation, and reduced task success, especially for temporal behaviors such as speed and smoothness.
We propose CTRL-STEER, a closed-loop framework that replaces static intervention strength with adaptive, time-varying control signals. The key idea is to decouple representation from regulation: rather than assuming temporal concepts are directly controlled by individual neurons, we steer along motion-aligned residual directions while a feedback controller adjusts intervention magnitude online. We instantiate this framework with both PID and reinforcement learning based controllers.
Experiments with a fine-tuned OpenVLA policy on four LIBERO task suites show that CTRL-STEER achieves more stable concept regulation and a better steering-task success trade-off than fixed-coefficient baselines, without modifying or retraining the base model.
Reader Mode unavailable (could not extract clean content).
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →MindGames Arena Generalization Track: In2AI Solution with Delayed Per-Step Reward Attribution
The In2AI solution introduces delayed per-step reward attribution for training language model agents in multi-agent environments, achieving top performance on the MindGames Arena benchmark at NeurIPS 2025. An 8-billion-parameter model outperformed larger proprietary systems, including GPT-5, in competitive play, demonstrating enhanced stability and sample efficiency in reinforcement learning.
