Finally, Section E illustrates qualitative results. We present the encoder-decoder variant of HAMT in fine-tuning on the right of Figure 1. Compared to the original cross-modal transformer on the left, the variant removes text-tovision cross-modal attention. The encoder encodes the texts to obtain textual embeddings. Theoriginal target location is viewed as a middle stop point.

HAMT efficientlyencodes allthepastpanoramic observationsviaahierarchical vision transformer (ViT), which first encodes individual images with ViT, then models spatial relation between images in a panoramic observation and finally takes into account temporal relation between panoramas in the history.

Developing robust and general-purpose manipulation policies represents a fundamental objective in robotics research. While Vision-Language-Action (VLA) models have demonstrated promising capabilities for end-to-end robot control, existing approaches still exhibit limited generalization to tasks beyond their training distributions. In contrast, humans possess remarkable proficiency in acquiring novel skills by simply observing others performing them once. Inspired by this capability, we propose ViVLA, a generalist robotic manipulation policy that achieves efficient task learning from a single expert demonstration video at test time. Our approach jointly processes an expert demonstration video alongside the robot's visual observations to predict both the demonstrated action sequences and subsequent robot actions, effectively distilling fine-grained manipulation knowledge from expert behavior and transferring it seamlessly to the agent. To enhance the performance of ViVLA, we develop a scalable expert-agent pair data generation pipeline capable of synthesizing paired trajectories from easily accessible human videos, further augmented by curated pairs from publicly available datasets. This pipeline produces a total of 892,911 expert-agent samples for training ViVLA. Experimental results demonstrate that our ViVLA is able to acquire novel manipulation skills from only a single expert demonstration video at test time. Our approach achieves over 30% improvement on unseen LIBERO tasks and maintains above 35% gains with cross-embodiment videos. Real-world experiments demonstrate effective learning from human videos, yielding more than 38% improvement on unseen tasks.

In recent years, Vision-Language-Action (VLA) models in embodied intelligence have developed rapidly. However, existing adversarial attack methods require costly end-to-end training and often generate noticeable perturbation patches. To address these limitations, we propose ADVLA, a framework that directly applies adversarial perturbations on features projected from the visual encoder into the textual feature space. ADVLA efficiently disrupts downstream action predictions under low-amplitude constraints, and attention guidance allows the perturbations to be both focused and sparse. We introduce three strategies that enhance sensitivity, enforce sparsity, and concentrate perturbations. Experiments demonstrate that under an $L_{\infty}=4/255$ constraint, ADVLA combined with Top-K masking modifies less than 10% of the patches while achieving an attack success rate of nearly 100%. The perturbations are concentrated on critical regions, remain almost imperceptible in the overall image, and a single-step iteration takes only about 0.06 seconds, significantly outperforming conventional patch-based attacks. In summary, ADVLA effectively weakens downstream action predictions of VLA models under low-amplitude and locally sparse conditions, avoiding the high training costs and conspicuous perturbations of traditional patch attacks, and demonstrates unique effectiveness and practical value for attacking VLA feature spaces.

Recent Vision-Language-Action (VLA) models have shown impressive flexibility and generalization, yet their deployment in robotic manipulation remains limited by heavy computational overhead and inference latency. In this work, we present ActDistill, a general action-guided self-derived distillation framework that transfers the action prediction capability of any existing VLA model to a lightweight counterpart. Unlike previous efficiency strategies that primarily emphasize vision-language correlations, ActDistill leverages action priors to guide knowledge transfer and model compression, achieving action-oriented efficiency for VLA models. Specifically, we employ a well-trained VLA model as the teacher and introduce a graph-structured encapsulation strategy to explicitly model the hierarchical evolution of action prediction. The student model, derived from the graph-encapsulated teacher, is further equipped with a dynamic router that adaptively selects computation paths based on action prediction demands, guided by hierarchical graph-informed supervision to ensure smooth and efficient evolution. During inference, graph-related auxiliary components are removed, allowing the student to execute only dynamically routed layers and predict high-precision actions with minimal computation and latency. Experiments on embodied benchmarks demonstrate that ActDistill achieves comparable or superior performance to full-scale VLA models while reducing computation by over 50% with up to 1.67 speedup, thereby establishing a general paradigm toward efficient embodied intelligence. Source codes can be found at https://github.com/

Recent vision-language-action (VLA) models for multi-task robotic manipulation commonly rely on static viewpoints and shared visual encoders, which limit 3D perception and cause task interference, hindering robustness and generalization. In this work, we propose Task-aware Virtual View Exploration (TVVE), a framework designed to overcome these challenges by integrating virtual view exploration with task-specific representation learning. TVVE employs an efficient exploration policy, accelerated by a novel pseudo-environment, to acquire informative views. Furthermore, we introduce a Task-aware Mixture-of-Experts (TaskMoE) visual encoder to disentangle features across different tasks, boosting both representation fidelity and task generalization. By learning to see the world in a task-aware way, TVVE generates more complete and discriminative visual representations, demonstrating significantly enhanced action prediction across a wide array of manipulation challenges. To further validate the robustness and generalization capability of TVVE under out-of-distribution (OOD) settings, we construct a challenging benchmark, RLBench-OG, covering various visual perturbations and camera pose variations. Extensive experiments on RLBench and RLBench-OG show that our TVVE achieves superior performance over state-of-the-art approaches. In real-robot experiments, TVVE demonstrates exceptional performance and generalizes robustly in multiple OOD settings, including visual disturbances and unseen instructions. Visual results and code are provided at: https://hcplab-sysu.github.io/TAVP.

Recent advances in Vision-Language-Action (VLA) models demonstrate that visual signals can effectively complement sparse action supervisions. However, letting VLA directly predict high-dimensional visual states can distribute model capacity and incur prohibitive training cost, while compressing visual states into more compact supervisory signals inevitably incurs information bottlenecks. Moreover, existing methods often suffer from poor comprehension and reasoning capabilities due to the neglect of language supervision. This paper introduces Mantis, a novel framework featuring a Disentangled Visual Foresight (DVF) to tackle these issues. Specifically, Mantis decouples visual foresight prediction from the backbone with the combination of meta queries and a diffusion Transformer (DiT) head. With the current visual state provided to the DiT via a residual connection, a simple next-state prediction objective enables the meta queries to automatically capture the latent actions that delineate the visual trajectory, and hence boost the learning of explicit actions. The disentanglement reduces the burden of the VLA backbone, enabling it to maintain comprehension and reasoning capabilities through language supervision. Empirically, pretrained on human manipulation videos, robot demonstrations, and image-text pairs, Mantis achieves a 96.7% success rate on LIBERO benchmark after fine-tuning, surpassing powerful baselines while exhibiting high convergence speed. Real-world evaluations show that Mantis outperforms $π_{0.5}$, a leading open-source VLA model, particularly in instruction-following capability, generalization to unseen instructions, and reasoning ability. Code and weights are released to support the open-source community.