# SphMind: Towards Robust, Training-Free VLM-based Spatial Reasoning with a 360 Camera

arXiv: https://arxiv.org/abs/2609.33462 · HTML: https://mlsanity.com/paper/2609.33462/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning.
- Builds on 34 of its 38 references that are in the map.
- First posted on arXiv 27 Sep 2026.

## Placed in

- LLM reasoning / [Spatial and embodied reasoning](https://mlsanity.com/lr/spatial-and-embodied-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)

## Builds on (most cited first, 30 of 34)

- [You Only Look Once: Unified, Real-Time Object Detection](https://mlsanity.com/paper/1506.02640/) (2015)
- [Evaluating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2305.10355/) (2023)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [Plug and Play Language Models: A Simple Approach to Controlled Text Generation](https://mlsanity.com/paper/1912.02164/) (2019)
- [Object Hallucination in Image Captioning](https://mlsanity.com/paper/1809.02156/) (2018)
- [Joint 2D-3D-Semantic Data for Indoor Scene Understanding](https://mlsanity.com/paper/1702.01105/) (2017)
- [Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs](https://mlsanity.com/paper/2206.11990/) (2022)
- [Learning SO(3) Equivariant Representations with Spherical CNNs](https://mlsanity.com/paper/1711.06721/) (2017)
- [UniFuse: Unidirectional Fusion for 360circ Panorama Depth Estimation](https://mlsanity.com/paper/2102.03550/) (2021)
- [Pano-AVQA: Grounded Audio-Visual Question Answering on 360^circ Videos](https://mlsanity.com/paper/2110.05122/) (2021)
- [Bending Reality: Distortion-aware Transformers for Adapting to Panoramic Semantic Segmentation](https://mlsanity.com/paper/2203.01452/) (2022)
- [Both Style and Distortion Matter: Dual-Path Unsupervised Domain Adaptation for Panoramic Semantic Segmentation](https://mlsanity.com/paper/2303.14360/) (2023)
- [Behind Every Domain There is a Shift: Adapting Distortion-aware Vision Transformers for Panoramic Semantic Segmentation](https://mlsanity.com/paper/2207.11860/) (2022)
- [Deep Learning for Omnidirectional Vision: A Survey and New Perspectives](https://mlsanity.com/paper/2205.10468/) (2022)
- [Look at the Neighbor: Distortion-aware Unsupervised Domain Adaptation for Panoramic Semantic Segmentation](https://mlsanity.com/paper/2308.05493/) (2023)
- [Semantics, Distortion, and Style Matter: Towards Source-free UDA for Panoramic Segmentation](https://mlsanity.com/paper/2403.12505/) (2024)
- [Elite360D: Towards Efficient 360 Depth Estimation via Semantic- and Distance-Aware Bi-Projection Fusion](https://mlsanity.com/paper/2403.16376/) (2024)
- [A Survey of Representation Learning, Optimization Strategies, and Applications for Omnidirectional Vision](https://mlsanity.com/paper/2502.10444/) (2025)
- [Omnidirectional CNN for Visual Place Recognition and Navigation](https://mlsanity.com/paper/1803.04228/) (2018)
- [Position: Beyond Euclidean -- Foundation Models Should Embrace Non-Euclidean Geometries](https://mlsanity.com/paper/2504.08896/) (2025)
- [Visual Question Answering on 360deg Images](https://mlsanity.com/paper/2001.03339/) (2020)
- [SphereUFormer: A U-Shaped Transformer for Spherical 360 Perception](https://mlsanity.com/paper/2412.06968/) (2024)
- [ODI-Bench: Can MLLMs Understand Immersive Omnidirectional Environments?](https://mlsanity.com/paper/2510.11549/) (2025)
- [Dream360: Diverse and Immersive Outdoor Virtual Scene Creation via Transformer-Based 360 Image Outpainting](https://mlsanity.com/paper/2401.10564/) (2024)
- [Beyond Flatlands: Unlocking Spatial Intelligence by Decoupling 3D Reasoning from Numerical Regression](https://mlsanity.com/paper/2511.11239/) (2025)
- [SpurLens: Automatic Detection of Spurious Cues in Multimodal LLMs](https://mlsanity.com/paper/2503.08884/) (2025)
- [PanoEnv: Exploring 3D Spatial Intelligence in Panoramic Environments with Reinforcement Learning](https://mlsanity.com/paper/2602.21992/) (2026)
- [LLMind: Bio-inspired Training-free Adaptive Visual Representations for Vision-Language Models](https://mlsanity.com/paper/2603.14882/) (2026)
- [GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs](https://mlsanity.com/paper/2507.18043/) (2025)
- [Learning to Ponder: Adaptive Reasoning in Latent Space](https://mlsanity.com/paper/2509.24238/) (2025)

## Abstract (arXiv)

Omnidirectional or 360 cameras provide embodied AI agents with a holistic, wide field-of-view (FoV) view of their surroundings, motivating the use of Multi-modal Large Language Models (MLLMs) for omnidirectional spatial reasoning. However, most MLLMs are trained on conventional 2D perspective images and struggle with the severe distortions and wrap-around discontinuities induced by spherical geometry. Enabling them to generalize to non-Euclidean 3D spaces without retraining therefore remains challenging. We propose SphMind, a training-free, plug-and-play framework that decouples semantic perception from geometric reasoning. Rather than requiring MLLMs to learn spherical geometry internally, SphMind preserves their semantic capabilities while handling geometry externally. We introduce a Spherical Harmonics-based Spatial Graph (SHSG) that models spatial relationships through equivariant transformations on the sphere, together with Inference-Time Geometric Grounding (IGG), a model-agnostic closed-loop optimization process that aligns MLLM representations with spherical geometric constraints during inference. Experiments on three benchmarks show that SphMind achieves over 21.4% average improvement in directional reasoning on MP3D and Stanford2D-3D, outperforms prompt-engineering baselines by 8.7% on the real-world ODI-Bench, and improves rotational invariance by 5.9% under panorama rotations, without additional training or dataset-specific tuning. In-the-wild evaluations further show that SphMind resolves directional reasoning queries that baseline vision-language models fail to answer correctly.