VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing

• URL: http://arxiv.org/abs/2510.05213v1
• Date: Mon, 06 Oct 2025 18:00:43 GMT
• Title: VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing
• Authors: Yixiao Wang, Mingxiao Huo, Zhixuan Liang, Yushi Du, Lingfeng Sun, Haotian Lin, Jinghuan Shang, Chensheng Peng, Mohit Bansal, Mingyu Ding, Masayoshi Tomizuka,
• Abstract summary: We propose VER, a Vision Expert transformer for Robot learning.<n>During pretraining, VER distills multiple VFMs into a vision expert library.<n>It then fine-tunes only a lightweight routing network to dynamically select task-relevant experts.
• Score: 89.48383845451717
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Pretrained vision foundation models (VFMs) advance robotic learning via rich visual representations, yet individual VFMs typically excel only in specific domains, limiting generality across tasks. Distilling multiple VFMs into a unified representation for policy can mitigate this limitation but often yields inflexible task-specific feature selection and requires costly full re-training to incorporate robot-domain knowledge. We propose VER, a Vision Expert transformer for Robot learning. During pretraining, VER distills multiple VFMs into a vision expert library. It then fine-tunes only a lightweight routing network (fewer than 0.4% of parameters) to dynamically select task-relevant experts from the pretrained library for downstream robot tasks. We further introduce Patchwise Expert Routing with Curriculum Top-K Annealing to improve both flexibility and precision of dynamic expert selection. Moreover, VER supports parameter-efficient finetuning for scalable expert utilization and adaptive robot-domain knowledge integration. Across 17 diverse robotic tasks and multiple policy heads, VER achieves state-of-the-art performance. We find that VER reduces large-norm outliers in task-irrelevant regions (e.g., background) and concentrates on task-critical regions. Visualizations and codes can be found in https://yixiaowang7.github.io/ver_page/.

Related papers

• UniVLA: Learning to Act Anywhere with Task-centric Latent Actions [32.83715417294052]
  UniVLA is a new framework for learning cross-embodiment vision-language-action (VLA) policies.<n>We derive task-centric action representations from videos with a latent action model.<n>We obtain state-of-the-art results across multiple manipulation and navigation benchmarks, as well as real-robot deployments.
  arXiv  Detail & Related papers  (2025-05-09T15:11:13Z)
• REMAC: Self-Reflective and Self-Evolving Multi-Agent Collaboration for Long-Horizon Robot Manipulation [57.628771707989166]
  We propose an adaptive multi-agent planning framework, termed REMAC, that enables efficient, scene-agnostic multi-robot long-horizon task planning and execution.<n>ReMAC incorporates two key modules: a self-reflection module performing pre-conditions and post-condition checks in the loop to evaluate progress and refine plans, and a self-evolvement module dynamically adapting plans based on scene-specific reasoning.
  arXiv  Detail & Related papers  (2025-03-28T03:51:40Z)
• Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers [41.069074375686164]
  We propose Heterogeneous Pre-trained Transformers (HPT), which pre-train a trunk of a policy neural network to learn a task and embodiment shared representation. We conduct experiments to investigate the scaling behaviors of training objectives, to the extent of 52 datasets. HPTs outperform several baselines and enhance the fine-tuned policy performance by over 20% on unseen tasks.
  arXiv  Detail & Related papers  (2024-09-30T17:39:41Z)
• Robotic Control via Embodied Chain-of-Thought Reasoning [86.6680905262442]
  Key limitation of learned robot control policies is their inability to generalize outside their training data.<n>Recent works on vision-language-action models (VLAs) have shown that the use of large, internet pre-trained vision-language models can substantially improve their robustness and generalization ability.<n>We introduce Embodied Chain-of-Thought Reasoning (ECoT) for VLAs, in which we train VLAs to perform multiple steps of reasoning about plans, sub-tasks, motions, and visually grounded features before predicting the robot action.
  arXiv  Detail & Related papers  (2024-07-11T17:31:01Z)
• Sparse Diffusion Policy: A Sparse, Reusable, and Flexible Policy for Robot Learning [61.294110816231886]
  We introduce a sparse, reusable, and flexible policy, Sparse Diffusion Policy (SDP) SDP selectively activates experts and skills, enabling efficient and task-specific learning without retraining the entire model. Demos and codes can be found in https://forrest-110.io/sparse_diffusion_policy/.
  arXiv  Detail & Related papers  (2024-07-01T17:59:56Z)
• LLaRA: Supercharging Robot Learning Data for Vision-Language Policy [56.505551117094534]
  We introduce LLaRA: Large Language and Robotics Assistant, a framework that formulates robot action policy as visuo-textual conversations.<n>First, we present an automated pipeline to generate conversation-style instruction tuning data for robots from existing behavior cloning datasets.<n>We show that a VLM finetuned with a limited amount of such datasets can produce meaningful action decisions for robotic control.
  arXiv  Detail & Related papers  (2024-06-28T17:59:12Z)
• LLARVA: Vision-Action Instruction Tuning Enhances Robot Learning [50.99807031490589]
  We introduce LLARVA, a model trained with a novel instruction tuning method to unify a range of robotic learning tasks, scenarios, and environments. We generate 8.5M image-visual trace pairs from the Open X-Embodiment dataset in order to pre-train our model. Experiments yield strong performance, demonstrating that LLARVA performs well compared to several contemporary baselines.
  arXiv  Detail & Related papers  (2024-06-17T17:55:29Z)
• Transferring Foundation Models for Generalizable Robotic Manipulation [82.12754319808197]
  We propose a novel paradigm that effectively leverages language-reasoning segmentation mask generated by internet-scale foundation models.<n>Our approach can effectively and robustly perceive object pose and enable sample-efficient generalization learning.<n>Demos can be found in our submitted video, and more comprehensive ones can be found in link1 or link2.
  arXiv  Detail & Related papers  (2023-06-09T07:22:12Z)
• PLEX: Making the Most of the Available Data for Robotic Manipulation Pretraining [28.504762473732296]
  We propose a transformer-based architecture that learns from a small amount of task-agnostic visuomotor trajectories. In particular, using relative positional encoding in plex's transformers greatly helps in low-data regimes of learning from human-collected demonstrations.
  arXiv  Detail & Related papers  (2023-03-15T17:31:37Z)
• PACT: Perception-Action Causal Transformer for Autoregressive Robotics Pre-Training [25.50131893785007]
  This work introduces a paradigm for pre-training a general purpose representation that can serve as a starting point for multiple tasks on a given robot. We present the Perception-Action Causal Transformer (PACT), a generative transformer-based architecture that aims to build representations directly from robot data in a self-supervised fashion. We show that finetuning small task-specific networks on top of the larger pretrained model results in significantly better performance compared to training a single model from scratch for all tasks simultaneously.
  arXiv  Detail & Related papers  (2022-09-22T16:20:17Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.