Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models

• URL: http://arxiv.org/abs/2606.02459v1
• Date: Mon, 01 Jun 2026 16:30:56 GMT
• Title: Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models
• Abstract summary: Existing approaches treat spatial language models (VLMs) as passive observers, which is difficult for real-world applications.<n>Inspired by pigeons' building and exploiting cognitive maps for navigation, we propose a novel agentic pipeline for spatial reasoning.<n>Experiments on VisionCube benchmark demonstrate state-of-the-art performance with emph80.5% overall accuracy, outperforming the best current method by emph29.5 accuracy points.
• Score: 17.24700274670692
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Enabling Vision-Language Models (VLMs) to perform spatial reasoning remains challenging. Existing approaches treat VLMs as passive observers, which is difficult for real-world applications. Moreover, reinforcement learning methods rely on sparse rewards, limiting their effectiveness for complex reasoning tasks. Inspired by pigeons' building and exploiting cognitive maps for navigation, we propose a novel agentic pipeline for spatial reasoning. First, we introduce a new \emph{dynamic cognitive map} parameterizing scene layout as object positions and orientations, serving as persistent memory for new observations. Second, we propose a novel \emph{Spatial Assertion Codes (SAC)}, Python expressions programmatically describing spatial relationships. By collaborating with the dynamic cognitive map, SAC enables verification of intermediate reasoning steps, providing dense reward signals. We optimize the model via supervised and reinforcement finetuning. Experiments on the MindCube benchmark demonstrate state-of-the-art performance with \emph{80.5\%} overall accuracy, outperforming the best current method by \emph{29.5} accuracy points (a relative improvement of \emph{53.2\%}) on the challenging \textsc{Rotation} subset. Our code and data are open-sourced at https://github.com/dw-dengwei/active-spatial-reasoning.git.

Related papers

• Plan in Sandbox, Navigate in Open Worlds: Learning Physics-Grounded Abstracted Experience for Embodied Navigation [63.11032720821731]
  textitSAGE is a framework that enables agents to learn within a physics-grounded semantic abstraction rather than a photorealistic simulation.<n>We demonstrate that textitSAGE significantly improves planner-assisted embodied navigation, achieving a 53.21% LLM-Match Success Rate on A-EQA.
  arXiv  Detail & Related papers  (2026-05-11T07:34:30Z)
• Entropy-Gradient Grounding: Training-Free Evidence Retrieval in Vision-Language Models [77.3748853516374]
  We propose a training-free, model-intrinsic grounding method for vision-language models.<n>We compute the entropy of the model's next-token distribution and backpropagate it to the visual token embeddings to obtain an entropy-gradient relevance map.<n>We then extract and rank multiple coherent regions to support multi-evidence queries, and introduce an iterative zoom-and-reground procedure with a spatial-entropy stopping rule.
  arXiv  Detail & Related papers  (2026-04-09T16:51:42Z)
• SignNav: Leveraging Signage for Semantic Visual Navigation in Large-Scale Indoor Environments [57.79171900005793]
  Humans routinely leverage semantic hints provided by signage to navigate to destinations within Large-Scale Indoor environments.<n>This paper introduces a novel embodied navigation task, SignNav, which requires the agent to interpret semantic hint from signage and reason about the subsequent action based on current observation.<n>We propose the spatial-aware module grounds the semantic hint of signage into physical world, while the temporal-aware module captures long-range dependencies between historical states and current observation.
  arXiv  Detail & Related papers  (2026-03-17T06:36:26Z)
• LLMs for Text-Based Exploration and Navigation Under Partial Observability [0.8478633389228728]
  Exploration and goal-directed navigation in unknown layouts are central to inspection, logistics, and search-and-rescue.<n>We ask whether large language models (LLMs) can function as emphtext-only controllers under partial observability.
  arXiv  Detail & Related papers  (2026-03-10T10:38:47Z)
• Boosting Zero-Shot VLN via Abstract Obstacle Map-Based Waypoint Prediction with TopoGraph-and-VisitInfo-Aware Prompting [18.325003967982827]
  Vision-language navigation (VLN) has emerged as a key task for embodied agents with broad practical applications.<n>We propose a zero-shot framework that integrates a simplified yet effective waypoint predictor with a multimodal large language model (MLLM)<n>Experiments on R2R-CE and RxR-CE show that our method achieves state-of-the-art zero-shot performance, with success rates of 41% and 36%, respectively.
  arXiv  Detail & Related papers  (2025-09-24T19:21:39Z)
• Self-Supervised Place Recognition by Refining Temporal and Featural Pseudo Labels from Panoramic Data [16.540900776820084]
  We propose a novel framework named TF-VPR that uses temporal neighborhoods and learnable feature neighborhoods to discover unknown spatial neighborhoods. Our method outperforms self-supervised baselines in recall rate, robustness, and heading diversity.
  arXiv  Detail & Related papers  (2022-08-19T12:59:46Z)
• Exploring Visual Context for Weakly Supervised Person Search [155.46727990750227]
  Person search has recently emerged as a challenging task that jointly addresses pedestrian detection and person re-identification. Existing approaches follow a fully supervised setting where both bounding box and identity annotations are available. This paper inventively considers weakly supervised person search with only bounding box annotations.
  arXiv  Detail & Related papers  (2021-06-19T14:47:13Z)
• MADE: Exploration via Maximizing Deviation from Explored Regions [48.49228309729319]
  In online reinforcement learning (RL), efficient exploration remains challenging in high-dimensional environments with sparse rewards. We propose a new exploration approach via textitmaximizing the deviation of the occupancy of the next policy from the explored regions. Our approach significantly improves sample efficiency over state-of-the-art methods.
  arXiv  Detail & Related papers  (2021-06-18T17:57:00Z)
• Latent World Models For Intrinsically Motivated Exploration [140.21871701134626]
  We present a self-supervised representation learning method for image-based observations. We consider episodic and life-long uncertainties to guide the exploration of partially observable environments.
  arXiv  Detail & Related papers  (2020-10-05T19:47:04Z)
• Learning Invariant Representations for Reinforcement Learning without Reconstruction [98.33235415273562]
  We study how representation learning can accelerate reinforcement learning from rich observations, such as images, without relying either on domain knowledge or pixel-reconstruction. Bisimulation metrics quantify behavioral similarity between states in continuous MDPs. We demonstrate the effectiveness of our method at disregarding task-irrelevant information using modified visual MuJoCo tasks.
  arXiv  Detail & Related papers  (2020-06-18T17:59:35Z)

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.