Modeling Dynamic Environments with Scene Graph Memory

TL;DR

This paper introduces Scene Graph Memory and a neural network architecture to predict object locations in dynamic, partially observable environments, improving search efficiency for embodied AI agents.

Contribution

It presents a novel scene graph memory representation and a Node Edge Predictor model for link prediction in dynamic, partially observable graphs, addressing a key challenge in embodied AI.

Findings

NEP outperforms baselines in diverse environments

SGM effectively captures accumulated observations

Method adapts well to various object movement dynamics

Abstract

Embodied AI agents that search for objects in large environments such as households often need to make efficient decisions by predicting object locations based on partial information. We pose this as a new type of link prediction problem: link prediction on partially observable dynamic graphs. Our graph is a representation of a scene in which rooms and objects are nodes, and their relationships are encoded in the edges; only parts of the changing graph are known to the agent at each timestep. This partial observability poses a challenge to existing link prediction approaches, which we address. We propose a novel state representation -- Scene Graph Memory (SGM) -- with captures the agent's accumulated set of observations, as well as a neural net architecture called a Node Edge Predictor (NEP) that extracts information from the SGM to search efficiently. We evaluate our method in the Dynamic House Simulator, a new benchmark that creates diverse dynamic graphs following the semantic patterns typically seen at homes, and show that NEP can be trained to predict the locations of objects in a variety of environments with diverse object movement dynamics, outperforming baselines both in terms of new scene adaptability and overall accuracy. The codebase and more can be found at https://www.scenegraphmemory.com.