$R^3$: Training Robots to Reason in Natural Language via Reinforcement Learning

Vision-Language-Action (VLA) models have gained much attention from the research community thanks to their strength in translating multimodal observations with linguistic instructions into desired robotic actions. Despite their advancements, VLAs often overlook explicit reasoning and learn the functional input-action mappings, omitting crucial logical steps, which are especially pronounced in interpretability and generalization for complex, long-horizon manipulation tasks. In this work, we propose ReFineVLA, a multimodal reasoning-aware framework that fine-tunes VLAs with teacher-guided reasons. We first augment robotic datasets with reasoning rationales generated by an expert teacher model, guiding VLA models to learn to reason about their actions. Then, we fine-tune pre-trained VLAs with the reasoning-enriched datasets with ReFineVLA, while maintaining the underlying generalization abilities and boosting reasoning capabilities. We also conduct attention map visualization to analyze the alignment among visual observation, linguistic prompts, and to-be-executed actions of ReFineVLA, reflecting the model is ability to focus on relevant tasks and actions. Through this additional step, we explore that ReFineVLA-trained models exhibit a meaningful agreement between vision-language and action domains, highlighting the enhanced multimodal understanding and generalization. Evaluated across a suite of simulated manipulation benchmarks on SimplerEnv with both WidowX and Google Robot tasks, ReFineVLA achieves state-of-the-art performance, in success rate over the second-best method on the both the WidowX benchmark and Google Robot Tasks.

Reinforcement learning (RL) for robotic manipulation often requires manually designing a dense reward function, which is difficult to tune and often fragile, or learning a reward from human demonstrations or preferences, which can be expensive. A recent line of work uses pretrained vision-language models (VLMs) as zero-shot reward models, replacing these costs with a single text prompt. However, we argue that a single global prompt is too coarse for long-horizon manipulation tasks with randomized initial conditions. The single-prompt VLM reward is near-flat for much of the trajectory, making early progress hard for the agent to detect. We propose Reinforced Micro-Task Learning (RMTL), an approach that decomposes a manipulation task into a small set of language-described micro-tasks and trains the agent to switch between them. At each step, the agent receives a multi-view VLM reward computed using the prompt of the currently active micro-task and averaged across multiple camera views to reduce the effect of view-specific occlusions. A reverse curriculum gradually exposes the agent to harder initial conditions, while a PPO worker is first trained with a fixed distance-based rule that selects the active micro-task. We then replace this rule with a learned hierarchical manager, turning rule-based phase selection into a fully learned hierarchical policy. We instantiate RMTL on the Fetch manipulation environment using three short stage-specific prompts and without additional prompt tuning. Experiments show that RMTL provides more informative reward signals than single-prompt VLM rewards, enabling faster learning. These results suggest that decomposing VLM rewards into micro-task-specific language prompts can substantially improve the scalability of language-guided reinforcement learning for robotic manipulation.

Embodied chain-of-thought (CoT) aims to bridge linguistic reasoning with robotic control, yet its effective form and integration remain underexplored. In this paper, we revisit embodied CoT for robotic control at an unprecedented scale. We curate the largest embodied CoT corpus to date, comprising 978,743 trajectories, 226.3M samples, and 2592.5 hours of data. Through extensive experiments, we show that effective CoT must ground high-level semantic understaning in concrete linguistic action guidance -- such as end-effector movement and image-space trajectories -- whereas high-level reasoning alone yields marginal gains. More importantly, we identify that explicit CoT does not scale reliably as an autoregressive action prefix, suffering from compounding errors during inference. To address these challenges, we propose ERVLA, a vision-language-action (VLA) model that effectively leverages linguistic reasoning in generalizable robot manipulation. ERVLA is trained using a CoT-dropout strategy, allowing the model to leverage rich reasoning traces during training while predicting actions directly without CoT during inference to bypass autoregressive instability. This approach enables reliable scaling with increasing pre-training data. ERVLA achieves state-of-the-art results on LIBERO-Plus with an 86.9% success rate and reaches 53.2% on VLABench, showcasing superior performance in out-of-distribution settings. Furthermore, ERVLA outperforms competitive state-of-the-art baselines in real-robot experiments, especially in handling semantic ambiguity and long-horizon tasks. Code, data, and model checkpoints are available on the project website: https://taoshuaiz.github.io/ERVLA/.