# Embodied Spatial Intelligence: from Implicit Scene Modeling to Spatial Reasoning

## TL;DR

This thesis advances embodied spatial intelligence by integrating implicit scene modeling and enhanced spatial reasoning in robots, enabling better perception and language-based decision-making in real-world environments.

## Contribution

It introduces novel scene representations and spatial reasoning methods that bridge language models and physical embodiment for robotic perception and action.

## Key findings

- Developed scalable implicit neural scene representations.
- Created a new navigation benchmark for spatial reasoning.
- Enhanced LLMs with grounding and feedback mechanisms.

## Abstract

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.