Direct Action-Head Injection of A Grounded 3D Point Unlocks Spatial and Task Generalization

• URL: http://arxiv.org/abs/2606.27663v1
• Date: Fri, 26 Jun 2026 02:44:27 GMT
• Title: Direct Action-Head Injection of A Grounded 3D Point Unlocks Spatial and Task Generalization
• Abstract summary: Vision-Language-Action (VLA) models leverage large-scale vision-language pretraining for flexible robot manipulation.<n>They remain brittle along two axes: spatial generalization, when object positions differ from those seen during training, and task generalization, when a familiar scene is paired with a different language instruction.<n>We propose a lightweight, model-agnostic module that represents the grounding signal in 3D, computes its relative displacement to the gripper, and injects the resulting embedding directly into the action head.
• Score: 6.3170928681089995
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Vision-Language-Action (VLA) models leverage large-scale vision-language pretraining for flexible robot manipulation, yet at test time they remain brittle along two axes: spatial generalization, when object positions differ from those seen during training, and task generalization, when a familiar scene is paired with a different language instruction than the one seen in training. A growing family of methods addresses this brittleness by endowing a policy with the spatial and task-aware information such as 2D pixel-coordinate for object localization and placement. However, we find that existing representation through language prompting or visual prompting does not address the limitations; in contrast, exploiting a 3D point-based representation and feeding it directly to the action head leads to substantial improvements-revealing that how the grounding signal is represented and injected into the VLA is the true game changer. Thus, we propose a lightweight, model-agnostic module that represents the grounding signal in 3D, computes its relative displacement to the gripper, and injects the resulting spatial embedding directly into the action head through adaptive layer normalization. The entire module is a two-layer MLP that requires no changes to the VLA backbone or pretraining pipeline. On LIBERO-PRO, our method improves the average success rate of GR00T-N1.6 from 31.2 to 77.5 points under task perturbation and from 28.1 to 60.2 points under position perturbation (gains of 46.3 and 32.1 points). Comparable gains are achieved for $π_{0.5}$ as well, demonstrating that the mechanism is backbone-agnostic. Together, these results support our central finding: given adequate grounding lifted into 3D, injecting it directly into the action head is what unlocks both spatial and task generalization in VLAs-achievable with nothing more than a lightweight module on top of a pretrained backbone.

Related papers

• Embodied Multimodal Grounding for Open-Vocabulary Mobile Manipulation via Semantic 3D Gaussian Splatting [4.473350810262674]
  Embodied mobile manipulation requires language, visual observations, three-dimensional scene structure, and action feasibility to be aligned before execution.<n>We study open-vocabulary target grounding with few-shot manipulation in local household workspaces and present an embodied multimodal grounding framework.<n>A task-driven local Semantic-3DGS serves as a shared interface across active sensing, language-conditioned 3D localization, obstacle-aware scene reasoning, base preparation, and semantic conditioning of the action model.
  arXiv  Detail & Related papers  (2026-08-11T10:16:30Z)
• Lift3D-VLA: Lifting VLA Models to 3D Geometry and Dynamics-Aware Manipulation [76.16065370700488]
  We introduce Lift3D-VLA, a unified VLA framework that equips models with explicit 3D point cloud reasoning and enables temporally coherent action generation.<n>We show that Lift3D-VLA achieves 10.8% and 11.1% higher mean success rates on MetaWorld and RLBench than the best-performing prior VLA methods.
  arXiv  Detail & Related papers  (2026-07-07T17:59:47Z)
• Sparse2Act: Learning Action-Aligned Sparse 3D Representations for Cross-Domain Robot Manipulation [52.00134393320209]
  We introduce Sparse2Act, an observation-action alignment framework for pretraining sparse point-cloud encoders.<n>Masked sparse 3D tokens are trained to organize scene features around the workspace motion paired with the observation.<n>On the LIBERO-10 benchmark, our method achieves 86.9% average success after 500 fine-tuning steps.
  arXiv  Detail & Related papers  (2026-06-10T23:56:01Z)
• 3DThinkVLA: Endowing Vision-Language-Action Models with Latent 3D Priors via 3D-Thinking-Guided Co-training [19.5808550016589]
  3D geometry perception and 3D spatial reasoning are distinct capabilities that can be disentangled and injected at different feature hierarchies.<n>We propose a 3D-thinking-guided co-training framework that enables vision-language-action (VLA) models to perform 3D spatial reasoning implicitly during action prediction.
  arXiv  Detail & Related papers  (2026-06-03T04:34:07Z)
• PointACT: Vision-Language-Action Models with Multi-Scale Point-Action Interaction [57.63073414949329]
  We propose PointACT, a dual-system 3D-aware VLA policy that integrates hierarchical 3D point cloud representations directly into the action decoding process.<n>PointACT employs a multi-scale point-action interaction mechanism with efficient bottleneck window self-attention, enabling evolving action tokens to densely attend to both local geometric detail and global scene structure.
  arXiv  Detail & Related papers  (2026-05-20T17:10:31Z)
• VEGA: Visual Encoder Grounding Alignment for Spatially-Aware Vision-Language-Action Models [55.12929235609365]
  Visual backbones of current vision-language-action (VLA) models are predominantly pretrained on 2D image data without explicit 3D geometric supervision.<n>Existing implicit grounding methods partially address this by aligning VLA features with those of 3D-aware foundation models.<n>We propose VEGA, a framework that directly aligns the output of the VLA's visual encoder with spatially-aware features from DINOv2-FiT3D.
  arXiv  Detail & Related papers  (2026-05-11T12:44:26Z)
• Universal Pose Pretraining for Generalizable Vision-Language-Action Policies [83.39008378156647]
  Existing Vision-Language-Action (VLA) models often suffer from feature collapse and low training efficiency.<n>We propose Pose-VLA, a decoupled paradigm that separates VLA training into a pre-training phase for extracting universal 3D spatial priors.<n>Our framework follows a two-stage pre-training pipeline, establishing fundamental spatial grounding via poses followed by motion alignment.
  arXiv  Detail & Related papers  (2026-02-23T11:00:08Z)
• ST4VLA: Spatially Guided Training for Vision-Language-Action Models [80.35847468618276]
  Large vision-language models (VLMs) excel at multimodal understanding but fall short when extended to embodied tasks.<n>We introduce ST4VLA, a dual-system Vision-Language-Action framework to align action learning with spatial priors inVLMs.<n> Empirically, ST4VLA achieves substantial improvements over vanilla VLA, with performance increasing from 66.1 -> 84.6 on Google Robot and from 54.7 -> 73.2 on WidowX Robot.
  arXiv  Detail & Related papers  (2026-02-10T18:59:17Z)
• Affordance Field Intervention: Enabling VLAs to Escape Memory Traps in Robotic Manipulation [25.496791183299692]
  Affordance Field Intervention (AFI) is a lightweight hybrid framework that uses SAFs as an on-demand plug-in to guide VLA behavior.<n>Our method achieves an average improvement of 23.5% across different VLA backbones.
  arXiv  Detail & Related papers  (2025-12-08T11:57:13Z)
• PointVLA: Injecting the 3D World into Vision-Language-Action Models [10.758939578236582]
  We propose PointVLA, a framework that enhances pre-trained vision-language-action models with point cloud inputs without requiring retraining.<n>Our method freezes the vanilla action expert and injects 3D features via a lightweight modular block.<n>PointVLA outperforms state-of-the-art 2D imitation learning methods across both simulated and real-world robotic tasks.
  arXiv  Detail & Related papers  (2025-03-10T16:32:41Z)
• LocATe: End-to-end Localization of Actions in 3D with Transformers [91.28982770522329]
  LocATe is an end-to-end approach that jointly localizes and recognizes actions in a 3D sequence. Unlike transformer-based object-detection and classification models which consider image or patch features as input, LocATe's transformer model is capable of capturing long-term correlations between actions in a sequence. We introduce a new, challenging, and more realistic benchmark dataset, BABEL-TAL-20 (BT20), where the performance of state-of-the-art methods is significantly worse.
  arXiv  Detail & Related papers  (2022-03-21T03:35:32Z)

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.