Situated Mapping of Sequential Instructions to Actions with Single-step Reward Observation

TL;DR

This paper introduces a novel attention-based learning approach for mapping sequential instructions to actions, utilizing single-step reward observations to improve accuracy in discourse-dependent tasks.

Contribution

It presents SESTRA, a new learning algorithm that leverages immediate reward signals for training without demonstrations, enhancing instruction-to-action mapping.

Findings

Achieved 9.8%-25.3% accuracy improvements on SCONE domains.

Effectively handles discourse and state dependencies in instruction following.

Outperforms high-level logical representation approaches.

Abstract

We propose a learning approach for mapping context-dependent sequential instructions to actions. We address the problem of discourse and state dependencies with an attention-based model that considers both the history of the interaction and the state of the world. To train from start and goal states without access to demonstrations, we propose SESTRA, a learning algorithm that takes advantage of single-step reward observations and immediate expected reward maximization. We evaluate on the SCONE domains, and show absolute accuracy improvements of 9.8%-25.3% across the domains over approaches that use high-level logical representations.