Factorizing Perception and Policy for Interactive Instruction Following

TL;DR

This paper introduces MOCA, a modular object-centric model that separates perception and policy to improve interactive instruction following in household tasks, demonstrating superior performance and generalization on the ALFRED benchmark.

Contribution

The paper proposes MOCA, a novel factorized model that enhances perception and policy streams for better interactive instruction following in AI agents.

Findings

MOCA outperforms prior methods on ALFRED benchmark.

MOCA shows improved generalization capabilities.

Empirical validation confirms the effectiveness of the factorized approach.

Abstract

Performing simple household tasks based on language directives is very natural to humans, yet it remains an open challenge for AI agents. The 'interactive instruction following' task attempts to make progress towards building agents that jointly navigate, interact, and reason in the environment at every step. To address the multifaceted problem, we propose a model that factorizes the task into interactive perception and action policy streams with enhanced components and name it as MOCA, a Modular Object-Centric Approach. We empirically validate that MOCA outperforms prior arts by significant margins on the ALFRED benchmark with improved generalization.