Towards Embodied Cognition in Robots via Spatially Grounded Synthetic Worlds

TL;DR

This paper introduces a synthetic dataset for training vision-language models to perform spatial reasoning, specifically focusing on inferring object distances along the Z-axis, as a step toward embodied cognition in robots.

Contribution

It presents a novel synthetic dataset generated in NVIDIA Omniverse for supervised learning of spatial reasoning tasks relevant to embodied AI.

Findings

Dataset enables supervised learning of spatial reasoning

Focus on inferring Z-axis distance as a foundational skill

Publicly available dataset supports further research

Abstract

We present a conceptual framework for training Vision-Language Models (VLMs) to perform Visual Perspective Taking (VPT), a core capability for embodied cognition essential for Human-Robot Interaction (HRI). As a first step toward this goal, we introduce a synthetic dataset, generated in NVIDIA Omniverse, that enables supervised learning for spatial reasoning tasks. Each instance includes an RGB image, a natural language description, and a ground-truth 4X4 transformation matrix representing object pose. We focus on inferring Z-axis distance as a foundational skill, with future extensions targeting full 6 Degrees Of Freedom (DOFs) reasoning. The dataset is publicly available to support further research. This work serves as a foundational step toward embodied AI systems capable of spatial understanding in interactive human-robot scenarios.