This thesis introduces "Embodied Spatial Intelligence" to address the challenge of creating robots that can perceive and act in the real world based on natural language instructions. To bridge the gap between Large Language Models (LLMs) and physical embodiment, we present contributions on two fronts: scene representation and spatial reasoning. For perception, we develop robust, scalable, and accurate scene representations using implicit neural models, with contributions in self-supervised camera calibration, high-fidelity depth field generation, and large-scale reconstruction. For spatial reasoning, we enhance the spatial capabilities of LLMs by introducing a novel navigation benchmark, a method for grounding language in 3D, and a state-feedback mechanism to improve long-horizon decision-making. This work lays a foundation for robots that can robustly perceive their surroundings and intelligently act upon complex, language-based commands.

Achieving human-like reasoning in deep learning models for complex tasks in unknown environments remains a critical challenge in embodied intelligence. While advanced vision-language models (VLMs) excel in static scene understanding, their limitations in spatio-temporal reasoning and adaptation to dynamic, open-set tasks like task-oriented navigation and embodied question answering (EQA) persist due to inadequate modeling of fine-grained spatio-temporal cues and physical world comprehension. To address this, we propose VEME, a novel cross-modal alignment method that enhances generalization in unseen scenes by learning an ego-centric, experience-centered world model. Our framework integrates three key components: (1) a cross-modal alignment framework bridging objects, spatial representations, and visual semantics with spatio-temporal cues to enhance VLM in-context learning; (2) a dynamic, implicit cognitive map activated by world embedding to enable task-relevant geometric-semantic memory recall; and (3) an instruction-based navigation and reasoning framework leveraging embodied priors for long-term planning and efficient exploration. By embedding geometry-aware spatio-temporal episodic experiences, our method significantly improves reasoning and planning in dynamic environments. Experimental results on VSI-Bench and VLN-CE demonstrate 1%-3% accuracy and exploration efficiency improvement compared to traditional approaches.

To localize, it matches these cues to a large-scale, indoor-outdoor navigation graph, constructed from publicly available maps. Abstract -- Navigation signs and maps, such as floor plans and street maps, are widely available and serve as ubiquitous aids for way-finding in human environments. Y et, they are rarely used by robot systems. This paper presents SignLoc, a global localization method that leverages navigation signs to localize the robot on publicly available maps--specifically floor plans and OpenStreetMap (OSM) graphs-without prior sensor-based mapping. It then employs a probabilistic observation model to match directional and locational cues from the detected signs to the graph, enabling robust topo-semantic localization within a Monte Carlo framework. We evaluated SignLoc in diverse large-scale environments: part of a university campus, a shopping mall, and a hospital complex. Experimental results show that SignLoc reliably localizes the robot after observing only one to two signs. Localizing and navigating in the open world remains a challenge for robots due to the diversity and complexity of human environments.

Recently, deep representation learning has shown strong performance in multiple audio tasks. However, its use for learning spatial representations from multichannel audio is underexplored. We investigate the use of a pretraining stage based on feature distillation to learn a robust spatial representation of binaural speech without the need for data labels. In this framework, spatial features are computed from clean binaural speech samples to form prediction labels. These clean features are then predicted from corresponding augmented speech using a neural network. After pretraining, we throw away the spatial feature predictor and use the learned encoder weights to initialize a DoA estimation model which we fine-tune for DoA estimation. Our experiments demonstrate that the pretrained models show improved performance in noisy and reverberant environments after fine-tuning for direction-of-arrival estimation, when compared to fully supervised models and classic signal processing methods.

Accurate sarcopenia diagnosis via ultrasound remains challenging due to subtle imaging cues, limited labeled data, and the absence of clinical context in most models. We propose MedVQA-TREE, a multimodal framework that integrates a hierarchical image interpretation module, a gated feature-level fusion mechanism, and a novel multi-hop, multi-query retrieval strategy. The vision module includes anatomical classification, region segmentation, and graph-based spatial reasoning to capture coarse, mid-level, and fine-grained structures. A gated fusion mechanism selectively integrates visual features with textual queries, while clinical knowledge is retrieved through a UMLS-guided pipeline accessing PubMed and a sarcopenia-specific external knowledge base. MedVQA-TREE was trained and evaluated on two public MedVQA datasets (VQA-RAD and PathVQA) and a custom sarcopenia ultrasound dataset. The model achieved up to 99% diagnostic accuracy and outperformed previous state-of-the-art methods by over 10%. These results underscore the benefit of combining structured visual understanding with guided knowledge retrieval for effective AI-assisted diagnosis in sarcopenia.

For human cognitive process, spatial reasoning and perception are closely entangled, yet the nature of this interplay remains underexplored in the evaluation of multimodal large language models (MLLMs). While recent MLLM advancements show impressive performance on reasoning, their capacity for human-like spatial cognition remains an open question. In this work, we introduce a systematic evaluation framework to assess the spatial reasoning abilities of state-of-the-art MLLMs relative to human performance. Central to our work is 11Plus-Bench, a high-quality benchmark derived from realistic standardized spatial aptitude tests. 11Plus-Bench also features fine-grained expert annotations of both perceptual complexity and reasoning process, enabling detailed instance-level analysis of model behavior. Through extensive experiments across 14 MLLMs and human evaluation, we find that current MLLMs exhibit early signs of spatial cognition. Despite a large performance gap compared to humans, MLLMs' cognitive profiles resemble those of humans in that cognitive effort correlates strongly with reasoning-related complexity. However, instance-level performance in MLLMs remains largely random, whereas human correctness is highly predictable and shaped by abstract pattern complexity. These findings highlight both emerging capabilities and limitations in current MLLMs' spatial reasoning capabilities and provide actionable insights for advancing model design.

Spatial representation learning is essential for GeoAI applications such as urban analytics, enabling the encoding of shapes, locations, and spatial relationships (topological and distance-based) of geo-entities like points, polylines, and polygons. Existing methods either target a single geo-entity type or, like Poly2Vec, decompose entities into simpler components to enable Fourier transformation, introducing high computational cost. Moreover, since the transformed space lacks geometric alignment, these methods rely on uniform, non-adaptive sampling, which blurs fine-grained features like edges and boundaries. To address these limitations, we introduce Geo2Vec, a novel method inspired by signed distance fields (SDF) that operates directly in the original space. Geo2Vec adaptively samples points and encodes their signed distances (positive outside, negative inside), capturing geometry without decomposition. A neural network trained to approximate the SDF produces compact, geometry-aware, and unified representations for all geo-entity types. Additionally, we propose a rotation-invariant positional encoding to model high-frequency spatial variations and construct a structured and robust embedding space for downstream GeoAI models. Empirical results show that Geo2Vec consistently outperforms existing methods in representing shape and location, capturing topological and distance relationships, and achieving greater efficiency in real-world GeoAI applications. Code and Data can be found at: https://github.com/chuchen2017/GeoNeuralRepresentation.

Recent advances in vision-language-action (VLA) models have shown promise in integrating image generation with action prediction to improve generalization and reasoning in robot manipulation. However, existing methods are limited to challenging image-based forecasting, which suffers from redundant information and lacks comprehensive and critical world knowledge, including dynamic, spatial and semantic information. To address these limitations, we propose DreamVLA, a novel VLA framework that integrates comprehensive world knowledge forecasting to enable inverse dynamics modeling, thereby establishing a perception-prediction-action loop for manipulation tasks. Specifically, DreamVLA introduces a dynamic-region-guided world knowledge prediction, integrated with the spatial and semantic cues, which provide compact yet comprehensive representations for action planning. This design aligns with how humans interact with the world by first forming abstract multimodal reasoning chains before acting. To mitigate interference among the dynamic, spatial and semantic information during training, we adopt a block-wise structured attention mechanism that masks their mutual attention, preventing information leakage and keeping each representation clean and disentangled. Moreover, to model the conditional distribution over future actions, we employ a diffusion-based transformer that disentangles action representations from shared latent features. Extensive experiments on both real-world and simulation environments demonstrate that DreamVLA achieves 76.7% success rate on real robot tasks and 4.44 average length on the CALVIN ABC-D benchmarks.

Spatial reasoning is a core component of human cognition, enabling individuals to perceive, comprehend, and interact with the physical world. It relies on a nuanced understanding of spatial structures and inter-object relationships, serving as the foundation for complex reasoning and decision-making. To investigate whether current vision-language models (VLMs) exhibit similar capability, we introduce Jigsaw-Puzzles, a novel benchmark consisting of 1,100 carefully curated real-world images with high spatial complexity. Based on this dataset, we design five tasks to rigorously evaluate VLMs' spatial perception, structural understanding, and reasoning capabilities, while deliberately minimizing reliance on domain-specific knowledge to better isolate and assess the general spatial reasoning capability. We conduct a comprehensive evaluation across 24 state-of-the-art VLMs. The results show that even the strongest model, Gemini-2.5-Pro, achieves only 77.14% overall accuracy and performs particularly poorly on the Order Generation task, with only 30.00% accuracy, far below the performance exceeding 90% achieved by human participants. This persistent gap underscores the need for continued progress, positioning Jigsaw-Puzzles as a challenging and diagnostic benchmark for advancing spatial reasoning research in VLMs. Our project page is at https://zesen01.github.io/jigsaw-puzzles.

Spatial cognition enables adaptive goal-directed behavior by constructing internal models of space. Robust biological systems consolidate spatial knowledge into three interconnected forms: \textit{landmarks} for salient cues, \textit{route knowledge} for movement trajectories, and \textit{survey knowledge} for map-like representations. While recent advances in multi-modal large language models (MLLMs) have enabled visual-language reasoning in embodied agents, these efforts lack structured spatial memory and instead operate reactively, limiting their generalization and adaptability in complex real-world environments. Here we present Brain-inspired Spatial Cognition for Navigation (BSC-Nav), a unified framework for constructing and leveraging structured spatial memory in embodied agents. BSC-Nav builds allocentric cognitive maps from egocentric trajectories and contextual cues, and dynamically retrieves spatial knowledge aligned with semantic goals. Integrated with powerful MLLMs, BSC-Nav achieves state-of-the-art efficacy and efficiency across diverse navigation tasks, demonstrates strong zero-shot generalization, and supports versatile embodied behaviors in the real physical world, offering a scalable and biologically grounded path toward general-purpose spatial intelligence.