RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics

• URL: http://arxiv.org/abs/2506.04308v3
• Date: Sat, 25 Oct 2025 03:07:10 GMT
• Title: RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics
• Authors: Enshen Zhou, Jingkun An, Cheng Chi, Yi Han, Shanyu Rong, Chi Zhang, Pengwei Wang, Zhongyuan Wang, Tiejun Huang, Lu Sheng, Shanghang Zhang,
• Abstract summary: Spatial referring is a fundamental capability of embodied robots to interact with the 3D physical world.<n>We propose RoboRefer, a 3D-aware VLM that can first achieve precise spatial understanding.<n>RoboRefer advances generalized multi-step spatial reasoning via reinforcement fine-tuning.
• Score: 67.11221574129937
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Spatial referring is a fundamental capability of embodied robots to interact with the 3D physical world. However, even with the powerful pretrained vision language models (VLMs), recent approaches are still not qualified to accurately understand the complex 3D scenes and dynamically reason about the instruction-indicated locations for interaction. To this end, we propose RoboRefer, a 3D-aware VLM that can first achieve precise spatial understanding by integrating a disentangled but dedicated depth encoder via supervised fine-tuning (SFT). Moreover, RoboRefer advances generalized multi-step spatial reasoning via reinforcement fine-tuning (RFT), with metric-sensitive process reward functions tailored for spatial referring tasks. To support SFT and RFT training, we introduce RefSpatial, a large-scale dataset of 20M QA pairs (2x prior), covering 31 spatial relations (vs. 15 prior) and supporting complex reasoning processes (up to 5 steps). In addition, we introduce RefSpatial-Bench, a challenging benchmark filling the gap in evaluating spatial referring with multi-step reasoning. Experiments show that SFT-trained RoboRefer achieves state-of-the-art spatial understanding, with an average success rate of 89.6%. RFT-trained RoboRefer further outperforms all other baselines by a large margin, even surpassing Gemini-2.5-Pro by 17.4% in average accuracy on RefSpatial-Bench. Notably, RoboRefer can be integrated with various control policies to execute long-horizon, dynamic tasks across diverse robots (e,g., UR5, G1 humanoid) in cluttered real-world scenes. Please see the project page at https://zhoues.github.io/RoboRefer.

Related papers

• pySpatial: Generating 3D Visual Programs for Zero-Shot Spatial Reasoning [18.697914587954163]
  pySpatial is a visual programming framework that equips MLLMs with the ability to interface with spatial tools.<n> pySpatial converts raw 2D inputs into an explorable 3D scene, enabling MLLMs to reason explicitly over structured spatial representations.
  arXiv  Detail & Related papers  (2026-03-01T03:55:49Z)
• Generalizable Geometric Prior and Recurrent Spiking Feature Learning for Humanoid Robot Manipulation [90.90219129619344]
  This paper presents a novel R-prior-S, Recurrent Geometric-priormodal Policy with Spiking features.<n>To ground high-level reasoning in physical reality, we leverage lightweight 2D geometric inductive biases.<n>For the data efficiency issue in robotic action generation, we introduce a Recursive Adaptive Spiking Network.
  arXiv  Detail & Related papers  (2026-01-13T23:36:30Z)
• RoboTracer: Mastering Spatial Trace with Reasoning in Vision-Language Models for Robotics [53.053660003572965]
  We propose RoboTracer, a 3D-aware VLM that first achieves both 3D spatial referring and measuring.<n>RoboTracer advances multi-step metric-grounded reasoning via reinforcement fine-tuning.<n>We present TraceSpatial-Bench, a challenging benchmark to evaluate spatial tracing.
  arXiv  Detail & Related papers  (2025-12-15T18:52:43Z)
• SpatialActor: Exploring Disentangled Spatial Representations for Robust Robotic Manipulation [63.48859753472547]
  SpatialActor is a framework for robust robotic manipulation that explicitly decouples semantics and geometry.<n>It achieves state-of-the-art performance with 87.4% on RLBench and improves by 13.9% to 19.4% under varying noisy conditions.
  arXiv  Detail & Related papers  (2025-11-12T18:59:08Z)
• InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy [138.89177083578213]
  We introduce InternVLA-M1, a unified framework for spatial grounding and robot control.<n>InternVLA-M1 employs a two-stage pipeline: (i) spatial grounding pre-training on over 2.3M spatial reasoning data, and (ii) spatially guided action post-training.<n>Results: InternVLA-M1 outperforms its variant without spatial guidance by +14.6% on SimplerEnv Google Robot, +17% on WidowX, and +4.3% on LIBERO Franka.
  arXiv  Detail & Related papers  (2025-10-15T17:30:05Z)
• SURPRISE3D: A Dataset for Spatial Understanding and Reasoning in Complex 3D Scenes [105.8644620467576]
  We introduce Stextscurprise3D, a novel dataset designed to evaluate language-guided spatial reasoning segmentation in complex 3D scenes.<n>Stextscurprise3D consists of more than 200k vision language pairs across 900+ detailed indoor scenes from ScanNet++ v2.<n>The dataset contains 89k+ human-annotated spatial queries deliberately crafted without object name.
  arXiv  Detail & Related papers  (2025-07-10T14:01:24Z)
• AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation [8.603450327406879]
  AnchorDP3 is a diffusion policy framework for dual-arm robotic manipulation.<n>It is trained on large-scale, procedurally generated simulation data.<n>It achieves a 98.7% average success rate in the RoboTwin benchmark.
  arXiv  Detail & Related papers  (2025-06-24T03:03:26Z)
• MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation [28.75149480374178]
  MEgoHand is a framework that synthesizes physically plausible hand-object interactions from egocentric RGB, text, and initial hand pose.<n>It achieves substantial reductions in wrist translation error and joint rotation error, highlighting its capacity to accurately model fine-grained hand joint structures.
  arXiv  Detail & Related papers  (2025-05-22T12:37:47Z)
• Robo2VLM: Visual Question Answering from Large-Scale In-the-Wild Robot Manipulation Datasets [16.46722805252982]
  Vision-Language Models (VLMs) acquire real-world knowledge and general reasoning ability through Internet-scale image-text corpora.<n>We present Robo2VLM, a Visual Question Answering (VQA) dataset generation framework for VLMs.<n>We curate Robo2VLM-1, a large-scale in-the-wild dataset with 684,710 questions covering 463 distinct scenes and 3,396 robotic manipulation tasks from 176k real robot trajectories.
  arXiv  Detail & Related papers  (2025-05-21T13:42:52Z)
• econSG: Efficient and Multi-view Consistent Open-Vocabulary 3D Semantic Gaussians [56.85804719947]
  We propose econSG for open-vocabulary semantic segmentation with 3DGS.<n>Our econSG shows state-of-the-art performance on four benchmark datasets compared to the existing methods.
  arXiv  Detail & Related papers  (2025-04-08T13:12:31Z)
• SpaceR: Reinforcing MLLMs in Video Spatial Reasoning [70.7401015322983]
  Video spatial reasoning poses a significant challenge for existing Multimodal Large Language Models (MLLMs)<n>This limitation stems primarily from 1) the absence of high-quality datasets for this task, and 2) the lack of effective training strategies to develop spatial reasoning capabilities.<n>Motivated by the success of Reinforcement Learning with Verifiable Reward (RLVR) in unlocking spatial reasoning abilities, this aims to improve MLLMs in video spatial reasoning through the RLVR paradigm.
  arXiv  Detail & Related papers  (2025-04-02T15:12:17Z)
• RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics [26.42651735582044]
  We introduce RoboSpatial, a large-scale dataset for spatial understanding in robotics.<n>It consists of real indoor and tabletop scenes, captured as 3D scans and egocentric images, and annotated with rich spatial information relevant to robotics.<n>Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robot manipulation.
  arXiv  Detail & Related papers  (2024-11-25T16:21:34Z)
• HiFi-CS: Towards Open Vocabulary Visual Grounding For Robotic Grasping Using Vision-Language Models [20.412403432289583]
  This paper introduces HiFi-CS, featuring hierarchical application of Featurewise Linear Modulation (FiLM) to fuse image and text embeddings.<n>Visual grounding associates an object in 2D/3D space with natural language input and is studied in two scenarios: Closed and Open Vocabulary.<n>We validate our approach through real-world RGS experiments using a 7-DOF robotic arm, achieving 90.33% visual grounding accuracy in 15 tabletop scenes.
  arXiv  Detail & Related papers  (2024-09-16T15:50:39Z)
• RoboSense: Large-scale Dataset and Benchmark for Egocentric Robot Perception and Navigation in Crowded and Unstructured Environments [62.5830455357187]
  We setup an egocentric multi-sensor data collection platform based on 3 main types of sensors (Camera, LiDAR and Fisheye)<n>A large-scale multimodal dataset is constructed, named RoboSense, to facilitate egocentric robot perception.
  arXiv  Detail & Related papers  (2024-08-28T03:17:40Z)
• GOI: Find 3D Gaussians of Interest with an Optimizable Open-vocabulary Semantic-space Hyperplane [53.388937705785025]
  3D open-vocabulary scene understanding is crucial for advancing augmented reality and robotic applications. We introduce GOI, a framework that integrates semantic features from 2D vision-language foundation models into 3D Gaussian Splatting (3DGS) Our method treats the feature selection process as a hyperplane division within the feature space, retaining only features that are highly relevant to the query.
  arXiv  Detail & Related papers  (2024-05-27T18:57:18Z)
• A Universal Semantic-Geometric Representation for Robotic Manipulation [42.18087956844491]
  We present $textbfSemantic-Geometric Representation (textbfSGR)$, a universal perception module for robotics. SGR leverages the rich semantic information of large-scale pre-trained 2D models and inherits the merits of 3D spatial reasoning. Our experiments demonstrate that SGR empowers the agent to successfully complete a diverse range of simulated and real-world robotic manipulation tasks.
  arXiv  Detail & Related papers  (2023-06-18T04:34:17Z)

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.