Learning Action Priors for Cross-embodiment Robot Manipulation

• URL: http://arxiv.org/abs/2606.26095v1
• Date: Wed, 24 Jun 2026 17:59:56 GMT
• Title: Learning Action Priors for Cross-embodiment Robot Manipulation
• Abstract summary: Most Vision-Language-Action (VLA) models build on a Vision-Language Model (VLM) backbone by attaching an action module and optimizing the full policy jointly.<n>This work proposes to pretrain the action module with motion priors before cross-modal VLA alignment.<n>We introduce a two-stage training framework that equips the action module with cross-embodiment temporal motion structure.
• Score: 46.92215687389683
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Most Vision-Language-Action (VLA) models build on a Vision-Language Model (VLM) backbone by attaching an action module and optimizing the full policy jointly. This design inherits strong visual and linguistic priors from the VLM, but leaves the action module to learn physical motion almost from scratch. As a result, the policy lacks an explicit motion prior, forcing early optimization to simultaneously discover temporal action dynamics and cross-modal alignment, a challenge further amplified in cross-embodiment settings. In this work, we propose to pretrain the action module with motion priors before cross-modal VLA alignment. Specifically, we introduce a two-stage training framework that equips the action module with cross-embodiment temporal motion structure before VLA training begins. In Stage~1, a lightweight flow-matching-based encoder-decoder action module efficiently learns temporal motion structure solely from unconditioned action trajectories, without processing visual or language tokens. In Stage~2, this learned prior is transferred to VLA training through decoder reuse and early-stage latent distillation, aligning visual-language features with the action embedding space while still allowing end-to-end policy refinement. In addition, the trained encoder serves as a compact history compressor, summarizing state-action histories into a single temporal context token for history-aware modeling at negligible cost. Extensive experiments across 13 diverse cross-embodiment tasks on both simulated and real-world platforms validate the effectiveness of our approach. Compared with VLA training without action priors, our model achieves faster convergence, higher success rates, and substantially stronger performance on data-scarce real-world tasks. Moreover, scaling up the action data in Stage~1 yields a more generalizable action prior that directly improves downstream VLA performance.

Related papers

• TS-Mask VLA: 2D Temporal-Spatial Masking for Vision-Language-Action Model with Effective Bridging [18.666233973595762]
  TS-Mask VLA is a vision-language-action framework for robot manipulation.<n>It is built upon two key designs: (1) a Discrete Action Expert equipped with a Bridge Attention conditioning bridge and (2) a temporal-spatial 2D masking strategy.<n>It achieves a 9 percent average success rate with only 0.5B parameters, outperforming significantly larger models.
  arXiv  Detail & Related papers  (2026-07-10T07:06:02Z)
• Two Bridges, One Pathway: From VLMs to Generalizable VLAs with Embodied Trajectory-Coupled Data [93.7685703383343]
  Vision-language models (VLMs) are powerful general-purpose reasoners, yet converting them into robot control policies is surprisingly difficult.<n>We argue that this gap can be bridged gradually with the right intermediate data.<n>We show that mixing task-relevant out-of-distribution ETC data with a small amount of action data enables the model to generalize to novel visual-language conditions.
  arXiv  Detail & Related papers  (2026-06-07T08:57:51Z)
• SpaAct: Spatially-Activated Transition Learning with Curriculum Adaptation for Vision-Language Navigation [40.318940817058746]
  Vision-and-Language Navigation (VLN) aims to enable an embodied agent to follow natural-language instructions and navigate to a target location in unseen 3D environments.<n>We argue that adapting VLMs to VLN requires endowing them with two complementary capabilities for acquiring such awareness.<n>We propose SpaAct, a training framework that activates the dynamic spatial awareness in VLMs.
  arXiv  Detail & Related papers  (2026-04-30T09:09:40Z)
• DIAL: Decoupling Intent and Action via Latent World Modeling for End-to-End VLA [72.9197085473598]
  DIAL is a framework bridging high-level decision making and low-level motor execution through a differentiable latent intent bottleneck.<n>A VLM-based System-2 performs latent world modeling by latent visual foresight within the VLM's native feature space.<n>A lightweight System-1 policy then decodes this predicted intent together with the current observation into precise robot actions.
  arXiv  Detail & Related papers  (2026-03-31T15:02:27Z)
• UMO: Unified In-Context Learning Unlocks Motion Foundation Model Priors [78.85130555487432]
  We present UMO, a simple yet general unified formulation that casts diverse downstream tasks into compositions of atomic per-frame operations.<n>Specifically, UMO introduces three learnable frame-level meta-operation embeddings to specify per-frame intent and employs lightweight temporal fusion to inject in-context cues into the pretrained backbone.<n>Experiments demonstrate that UMO consistently outperforms task-specific and training-free baselines across a wide range of benchmarks.
  arXiv  Detail & Related papers  (2026-03-16T22:44:52Z)
• ConLA: Contrastive Latent Action Learning from Human Videos for Robotic Manipulation [27.54751123419347]
  We propose ConLA, an unsupervised pretraining framework for learning robotic policies from human videos.<n>By pretraining solely on human videos, our method for the first time surpasses the performance obtained with real robot trajectory pretraining.
  arXiv  Detail & Related papers  (2026-01-31T06:40:57Z)
• Robotic VLA Benefits from Joint Learning with Motion Image Diffusion [114.60268819583017]
  Vision-Language-Action (VLA) models have achieved remarkable progress in robotic manipulation by mapping multimodal observations and instructions directly to actions.<n>We propose joint learning with motion image diffusion, a novel strategy that enhances VLA models with motion reasoning capabilities.<n> Experiments in both simulation and real-world environments demonstrate that joint learning with motion image diffusion improves the success rate of pi-series VLAs to 97.5%.
  arXiv  Detail & Related papers  (2025-12-19T19:07:53Z)
• VITA-VLA: Efficiently Teaching Vision-Language Models to Act via Action Expert Distillation [76.13140980997508]
  Vision-Language Action (VLA) models significantly advance robotic manipulation by leveraging the strong perception capabilities of pretrained vision-language models (VLMs)<n>We propose a simple yet effective distillation-based framework that equips VLMs with action-execution capability by transferring knowledge from pretrained small action models.<n>In real-world experiments across five manipulation tasks, our method consistently outperforms the teacher model, achieving 82.0% success rate (17% improvement)
  arXiv  Detail & Related papers  (2025-10-10T17:59:56Z)

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.