Grounding Vision and Language to 3D Masks for Long-Horizon Box Rearrangement

We present ZeroDex, a zero-shot framework for long-horizon dexterous manipulation that grounds language instructions into executable 3D task plans from calibrated multi-view RGB images. Our framework builds on the grounding ability that vision-language models (VLMs) already exhibit in 2D and lifts their predictions into 3D through multi-view fusion, which extends VLM grounding beyond image-space keypoints to the metric 3D quantities that manipulation actually requires: contact points, placement targets, and tool-motion anchors. Because execution is driven by these 3D groundings rather than a learned policy, the system preserves the VLM's open-world generalization, without collecting demonstrations or post-training for a specific task or embodiment. The VLM decomposes a complex instruction into a sequence of atomic tasks, each executed by a primitive from a small reusable library that covers manipulation tasks including pick-and-place and tool-use: for tool-use, we retrieve an object-centric atomic action corresponding to the inferred skill category and align its stored 6D tool trajectory to the grounded keypoints; for dexterous execution, we expand the lifted grasp keypoint into a task-conditioned grasp affordance region and generate feasible grasp-motion pairs with an arm-hand motion generator. Real-world experiments show improved 3D grounding accuracy and higher task success than zero-shot visual-prompting and constraint-based baselines, both of which ground from a single RGB-D view, and our demonstration-free system outperforms VLA policies fine-tuned on 100 task demonstrations. We further demonstrate long-horizon manipulation, where execution verification and replanning recover from intermediate failures, and show that the system generalizes to unseen objects and tool-use tasks in novel scenes.

Vision-language models reliably name objects in a scene, but do they represent the 3D layout those objects inhabit? We introduce a 3,034-sample human-curated benchmark targeting three components of spatial understanding: depth-ordered occlusion (probed via three independent counterfactual operationalisations), optical-geometry inference over visible reflections, and volumetric rearrangement planning. Six frontier and open-weight VLMs, scored by trained annotators on 18,204 responses with no LLM-as-judge, reveal a sharp dissociation: models that plan rearrangements over visible layouts at 53--97% accuracy and rarely violate collision constraints fall to 6--45% on occlusion and below 7% on reflections. An embodied-reasoning model reproduces the same profile. White-box analysis on Qwen3-VL-8B-Thinking localises the failure to the visual-token merger: spatial information recoverable throughout the vision encoder becomes inaccessible after token compression and only stabilises again when clean post-merger activations are patched into the language decoder.

Vision-Language-Action (VLA) models have shown strong potential for general robot manipulation, but most existing models rely on 2D visual-language backbones and lack fine-grained 3D understanding of target objects, especially under occlusion, pose variation, scale changes, and precise spatial interaction. We propose an object-centric 3D representation alignment framework built upon $π_0$, using SAM3D as a frozen 3D teacher to provide target-object 3D priors during training. Specifically, we localize task-relevant objects with object recognition models, generate corresponding object masks, and use SAM3D to extract dense object-level 3D representations, which are aligned with intermediate visual features of $π_0$. This enables the policy to internalize target-object 3D information while preserving the original RGB-language-to-action inference pipeline without requiring depth, point clouds, masks, SAM3D, or additional 3D modules at test time. Simulation experiments show consistent improvements, achieving 99.1% on LIBERO and an average length of 4.11 on CALVIN. Real-world experiments further demonstrate that our method is particularly effective in long-horizon manipulation scenarios where the robot must focus on different target objects across multiple subtasks.