VLA 研究日報VLA 研究日报

Dexterous manipulation is essential for real-world robot autonomy, mirroring the central role of human hand coordination in daily activity. Humans rely on rich multimodal perception--vision, sound, and language-guided intent--to perform dexterous actions, motivating vision-based, language-conditioned manipulation systems for robots. However, training reliable vision-language-action (VLA) models for dexterous manipulation requires large-scale demonstrations across many robotic hands. In addition, as new dexterous embodiments appear rapidly, collecting data for each becomes costly and impractical, creating a need for scalable cross-embodiment learning. We introduce XL-VLA, a vision-language-action framework integrated with a unified latent action space shared across diverse dexterous hands. This embodiment-invariant latent space is directly pluggable into standard VLA architectures, enabling seamless cross-embodiment training and efficient reuse of both existing and newly collected data. Experimental results demonstrate that XL-VLA consistently outperforms baseline VLA models operating in raw joint spaces, establishing it as an effective solution for scalable cross-embodiment dexterous manipulation.

[Berkeley|Duan]

Vision-Language-Action (VLA) models enable generalist robotic manipulation but suffer from high inference latency. This bottleneck stems from the massive number of visual tokens processed by large language backbones. Existing methods either prune or merge tokens uniformly, degrading the spatial reasoning essential for robotic control. We present DepthCache, a training-free framework that leverages depth as a structural prior for visual token compression. It partitions observations into depth-based regions and applies spatially differentiated merge ratios, preserving the near-field workspace while compressing the distant background. To exploit temporal redundancy, DepthCache distributes the merging process across consecutive frames, ensuring consistent representations while reducing per-step computation. A motion-adaptive pipeline further optimizes auxiliary view compression based on end-effector dynamics. The framework requires no model modification, generalizing across diverse VLA architectures. On the LIBERO benchmark, DepthCache achieves up to 1.28x inference speedup with less than 1% average success rate degradation across three VLA models (pi_0.5, OpenVLA, GR00T), whereas pruning and merging baselines incur 4--24% degradation at comparable compression. Real-world experiments on a physical manipulator demonstrate that DepthCache enables faster task throughput and more responsive closed-loop control in latency-sensitive scenarios.

[北航]

Vision-Language-Action (VLA) models have emerged as a promising paradigm for robot learning, but their representations are still largely inherited from static image-text pretraining, leaving physical dynamics to be learned from comparatively limited action data. Generative video models, by contrast, encode rich spatiotemporal structure and implicit physics, making them a compelling foundation for robotic manipulation. But their potentials are not fully explored in the literature. To bridge the gap, we introduce DiT4DiT, an end-to-end Video-Action Model that couples a video Diffusion Transformer with an action Diffusion Transformer in a unified cascaded framework. Instead of relying on reconstructed future frames, DiT4DiT extracts intermediate denoising features from the video generation process and uses them as temporally grounded conditions for action prediction. We further propose a dual flow-matching objective with decoupled timesteps and noise scales for video prediction, hidden-state extraction, and action inference, enabling coherent joint training of both modules. Across simulation and real-world benchmarks, DiT4DiT achieves state-of-the-art results, reaching average success rates of 98.6% on LIBERO and 50.8% on RoboCasa GR1 while using substantially less training data. On the Unitree G1 robot, it also delivers superior real-world performance and strong zero-shot generalization. Importantly, DiT4DiT improves sample efficiency by over 10x and speeds up convergence by up to 7x, demonstrating that video generation can serve as an effective scaling proxy for robot policy learning. We release code and models at https://dit4dit.github.io/.

[科大]

Dexterous manipulation is limited by both control and design, without consensus as to what makes manipulators best for performing dexterous tasks. This raises a fundamental challenge: how should we design and control robot manipulators that are optimized for dexterity? We present a co-design framework that learns task-specific hand morphology and complementary dexterous control policies. The framework supports 1) an expansive morphology search space including joint, finger, and palm generation, 2) scalable evaluation across the wide design space via morphology-conditioned cross-embodied control, and 3) real-world fabrication with accessible components. We evaluate the approach across multiple dexterous tasks, including in-hand rotation with simulation and real deployment. Our framework enables an end-to-end pipeline that can design, train, fabricate, and deploy a new robotic hand in under 24 hours. The full framework will be open-sourced and available on our website: https://an-axolotl.github.io/HouseofDextra/ .

[UCSD|Su]

We introduce a lifelong imitation learning framework that enables continual policy refinement across sequential tasks under realistic memory and data constraints. Our approach departs from conventional experience replay by operating entirely in a multimodal latent space, where compact representations of visual, linguistic, and robot's state information are stored and reused to support future learning. To further stabilize adaptation, we introduce an incremental feature adjustment mechanism that regularizes the evolution of task embeddings through an angular margin constraint, preserving inter-task distinctiveness. Our method establishes a new state of the art in the LIBERO benchmarks, achieving 10-17 point gains in AUC and up to 65% less forgetting compared to previous leading methods. Ablation studies confirm the effectiveness of each component, showing consistent gains over alternative strategies. The code is available at: https://github.com/yfqi/lifelong_mlr_ifa.

[IIT]

Robotic cloth manipulation remains challenging due to the high-dimensional state space of fabrics, their deformable nature, and frequent occlusions that limit vision-based sensing. Although dual-arm systems can mitigate some of these issues, they increase hardware and control complexity. This paper presents Touch G.O.G., a compact vision-based tactile gripper and perception/control framework for single-arm bimanual cloth manipulation. The proposed framework combines three key components: (1) a novel gripper design and control strategy for in-gripper cloth sliding with a single robot arm, (2) a Vision Foundation Model-backboned Vision Transformer pipeline for cloth part classification (PC-Net) and edge pose estimation (PE-Net) using real and synthetic tactile images, and (3) an encoder-decoder synthetic data generator (SD-Net) that reduces manual annotation by producing high-fidelity tactile images. Experiments show 96% accuracy in distinguishing edges, corners, interior regions, and grasp failures, together with sub-millimeter edge localization and 4.5° orientation error. Real-world results demonstrate reliable cloth unfolding, even for crumpled fabrics, using only a single robotic arm. These results highlight Touch G.O.G. as a compact and cost-effective solution for deformable object manipulation.

[Imperial|Kormushev]

Latent Action Models (LAMs) enable Vision- Language-Action (VLA) systems to learn semantic action representations from large-scale unannotated data. Yet, we identify two bottlenecks of LAMs: 1) the commonly adopted end-to-end trained image encoder suffers from poor spatial understanding; 2) LAMs can be fragile when input frames are temporally distant, leading to limited temporal percep- tion. Such factors inevitably hinder stable and clear action modeling. To this end, we propose Farsighted-LAM, a latent action framework with geometry-aware spatial encoding and multi-scale temporal modeling, capturing structural priors and dynamic motion patterns from consecutive frames. We further propose SSM-VLA, an end-to-end VLA framework built upon Farsighted-LAM, which integrates structured perception with a visual Chain-of-Thought module to explicitly reason about environmental dynamics, enhancing decision consistency and interpretability. We validate SSM-VLA on multiple VLA tasks in both simulation and real-world settings, and achieve state-of- the-art performance. Our results demonstrate that our strategy of combining geometry-aware modeling, temporal coherence, and explicit reasoning is effective in enhancing the robustness and generalizability of embodied intelligence.

[上交]

World Models (WMs) have emerged as a promising approach for post-training Vision-Language-Action (VLA) policies to improve robustness and generalization under environmental changes. However, most WM-based post-training methods rely on pixel-space supervision, making policies sensitive to pixel-level artifacts and hallucination from imperfect WM rollouts. We introduce World2Act, a post-training framework that aligns VLA actions directly with WM video-dynamics latents using a contrastive matching objective, reducing dependence on pixels. Post-training performance is tied to rollout quality, yet current WMs struggle with arbitrary-length video generation as they are mostly trained on fixed-length clips while robotic execution durations vary widely. To address this, we propose an automatic LLM-based skill-decomposition pipeline that segments high-level instructions into low-level prompts. Our pipeline produces RoboCasa-Skill and LIBERO-Skill, supporting skill-compositional WMs that remain temporally consistent across diverse task horizons. Empirically, applying World2Act to VLAs like GR00T-N1.6 and Cosmos Policy achieves state-of-the-art results on RoboCasa and LIBERO, and improves real-world performance by 6.7%, enhancing embodied agent generalization.

[UGA|Liang]

Recent Vision-Language-Action (VLA) models reformulate vision-language models by tuning them with millions of robotic demonstrations. While they perform well when fine-tuned for a single embodiment or task family, extending them to multi-skill settings remains challenging: directly merging VLA experts trained on different tasks results in near-zero success rates. This raises a fundamental question: what prevents VLAs from mastering multiple skills within one model? With an empirical decomposition of learnable parameters during VLA fine-tuning, we identify two key sources of non-mergeability: (1) Finetuning drives LoRA adapters in the VLM backbone toward divergent, task-specific directions beyond the capacity of existing merging methods to unify. (2) Action experts develop inter-block dependencies through self-attention feedback, causing task information to spread across layers and preventing modular recombination. To address these challenges, we present MergeVLA, a merging-oriented VLA architecture that preserves mergeability by design. MergeVLA introduces sparsely activated LoRA adapters via task masks to retain consistent parameters and reduce irreconcilable conflicts in the VLM. Its action expert replaces self-attention with cross-attention-only blocks to keep specialization localized and composable. When the task is unknown, it uses a test-time task router to adaptively select the appropriate task mask and expert head from the initial observation, enabling unsupervised task inference. Across LIBERO, LIBERO-Plus, RoboTwin, and multi-task experiments on the real SO101 robotic arm, MergeVLA achieves performance comparable to or even exceeding individually finetuned experts, demonstrating robust generalization across tasks, embodiments, and environments. Project page: https://mergevla.github.io/

[UQ]

We present DAWN (Diffusion is All We Need for robot control), a unified diffusion-based framework for language-conditioned robotic manipulation that bridges high-level motion intent and low-level robot action via structured pixel motion representation. In DAWN, both the high-level and low-level controllers are modeled as diffusion processes, yielding a fully trainable, end-to-end system with interpretable intermediate motion abstractions. DAWN achieves state-of-the-art results on the challenging CALVIN benchmark, demonstrating strong multi-task performance, and further validates its effectiveness on MetaWorld. Despite the substantial domain gap between simulation and reality and limited real-world data, we demonstrate reliable real-world transfer with only minimal finetuning, illustrating the practical viability of diffusion-based motion abstractions for robotic control. Our results show the effectiveness of combining diffusion modeling with motion-centric representations as a strong baseline for scalable and robust robot learning. Project page: https://eronguyen.github.io/DAWN/

[DeepMind|Ryoo]

Recent advancements in integrating tactile sensing into vision-language-action (VLA) models have demonstrated transformative potential for robotic perception. However, existing tactile representations predominantly rely on qualitative descriptors (e.g., texture), neglecting quantitative contact states such as force magnitude, contact geometry, and principal axis orientation, which are indispensable for fine-grained manipulation. To bridge this gap, we propose FG-CLTP, a fine-grained contrastive language tactile pretraining framework. We first introduce a novel dataset comprising over 100k tactile 3D point cloud-language pairs that explicitly capture multidimensional contact states from the sensor's perspective. We then implement a discretized numerical tokenization mechanism to achieve quantitative-semantic alignment, effectively injecting explicit physical metrics into the multimodal feature space. The proposed FG-CLTP model yields a 95.9% classification accuracy and reduces the regression error (MAE) by 52.6% compared to state-of-the-art methods. Furthermore, the integration of 3D point cloud representations establishes a sensor-agnostic foundation with a minimal sim-to-real gap of 3.5%. Building upon this fine-grained representation, we develop a 3D tactile-language-action (3D-TLA) architecture driven by a flow matching policy to enable multimodal reasoning and control. Extensive experiments demonstrate that our framework significantly outperforms strong baselines in contact-rich manipulation tasks, providing a robust and generalizable foundation for tactile-language-action models.

[中科院]