ALOE: Action-Level Off-Policy Evaluation for Vision-Language-Action Model Post-Training

We study how to improve large foundation vision-language-action (VLA) systems through online reinforcement learning (RL) in real-world settings. Central to this process is the value function, which provides learning signals to guide VLA learning from experience. In practice, the value function is estimated from trajectory fragments collected from different data sources, including historical policies and intermittent human interventions. Estimating the value function of current behavior quality from the mixture data is inherently an off-policy evaluation problem. However, prior work often adopts conservative on-policy estimation for stability, which avoids direct evaluation of the current high-capacity policy and limits learning effectiveness. In this paper, we propose ALOE, an action-level off-policy evaluation framework for VLA post-training. ALOE applies chunking-based temporal-difference bootstrapping to evaluate individual action sequences instead of predicting final task outcomes. This design improves effective credit assignment to critical action chunks under sparse rewards and supports stable policy improvement. We evaluate our method on three real-world manipulation tasks, including smartphone packing as a high-precision task, laundry folding as a long-horizon deformable-object task, and bimanual pick-and-place involving multi-object perception. Across all tasks, ALOE improves learning efficiency without compromising execution speed, showing that off-policy RL can be reintroduced in a reliable manner for real-world VLA post-training. Videos and additional materials are available at our project website.

Hierarchical Visual Policy Learning for Long-Horizon Robot Manipulation in Densely Cluttered Scenes

In this work, we focus on addressing the long-horizon manipulation tasks in densely cluttered scenes. Such tasks require policies to effectively manage severe occlusions among objects and continually produce actions based on visual observations. We propose a vision-based Hierarchical policy for Cluttered-scene Long-horizon Manipulation (HCLM). It employs a high-level policy and three options to select and instantiate three parameterized action primitives: push, pick, and place. We first train the pick and place options by behavior cloning (BC). Subsequently, we use hierarchical reinforcement learning (HRL) to train the high-level policy and push option. During HRL, we propose a Spatially Extended Q-update (SEQ) to augment the updates for the push option and a Two-Stage Update Scheme (TSUS) to alleviate the non-stationary transition problem in updating the high-level policy. We demonstrate that HCLM significantly outperforms baseline methods in terms of success rate and efficiency in diverse tasks. We also highlight our method's ability to generalize to more cluttered environments with more additional blocks.