arXiv:2510.00600v2 Announce Type: replace-cross 
Abstract: Using Large Language Models to produce intermediate thoughts, a.k.a. Chain-of-thought (CoT), before providing an answer has been a successful recipe for solving complex language tasks. In robotics, similar embodied CoT strategies, generating thoughts before actions, have also been shown to lead to improved performance when using Vision-Language-Action models (VLAs). As these techniques increase the length of the model's generated outputs to include the thoughts, the inference time is negatively affected. Delaying an agent's actions in real-world executions, as in robotic manipulation settings, strongly affects the usability of a method, as tasks require long sequences of actions. However, is the generation of long chains-of-thought a strong prerequisite for achieving performance improvements? In this work, we explore the idea of Hybrid Training (HyT), a framework that enables VLAs to learn from thoughts and benefit from the associated performance gains, while enabling the possibility to leave out CoT generation during inference. Furthermore, by learning to conditionally predict a diverse set of outputs, HyT supports flexibility at inference time, enabling the model to either predict actions directly, generate thoughts or follow instructions. We evaluate the proposed method in a series of simulated benchmarks and real-world experiments.

أدت التطورات الأخيرة في الذكاء الاصطناعي إلى استكشاف التدريب الهجين (HyT) لنماذج العمل-اللغة-الرؤية (VLA)، والذي يهدف إلى تحسين الأداء من خلال السماح لهذه النماذج بالتعلم من الأفكار الوسيطة قبل تنفيذ الإجراءات. تتناول هذه الطريقة التحدي المتمثل في وقت الاستدلال، الذي يمكن أن يتأثر سلبًا بسلاسل الأفكار الأطول.

Los avances recientes en inteligencia artificial han llevado a la exploración del Entrenamiento Híbrido (HyT) para modelos de Acción-Lenguaje-Visión (VLA), que busca mejorar el rendimiento al permitir que estos modelos aprendan de pensamientos intermedios antes de ejecutar acciones. Este enfoque aborda el desafío del tiempo de inferencia, que puede verse negativamente afectado por cadenas de pensamiento más largas.

Des avancées récentes en intelligence artificielle ont conduit à l'exploration de l'Entraînement Hybride (HyT) pour les modèles Vision-Langage-Action (VLA), qui vise à améliorer les performances en permettant à ces modèles d'apprendre à partir de pensées intermédiaires avant d'exécuter des actions. Cette approche aborde le défi du temps d'inférence, qui peut être négativement impacté par des chaînes de pensée plus longues.

Recent advancements in artificial intelligence have led to the exploration of Hybrid Training (HyT) for Vision-Language-Action (VLA) models, which aims to enhance performance by allowing these models to learn from intermediate thoughts before executing actions. This approach addresses the challenge of inference time, which can be negatively impacted by longer chains of thought.