Improving Long-Horizon Imitation Through Instruction Prediction

TL;DR

This paper introduces an instruction prediction auxiliary task to improve long-horizon planning in imitation learning, especially in low-data regimes, by leveraging language to enhance high-level, temporally extended representations.

Contribution

It proposes using instruction prediction as an auxiliary loss to enhance planning performance in low-data settings, demonstrating significant improvements on BabyAI and Crafter benchmarks.

Findings

Instruction modeling boosts performance in complex planning tasks.

Most beneficial for tasks requiring complex reasoning.

Provides larger gains in low-data regimes.

Abstract

Complex, long-horizon planning and its combinatorial nature pose steep challenges for learning-based agents. Difficulties in such settings are exacerbated in low data regimes where over-fitting stifles generalization and compounding errors hurt accuracy. In this work, we explore the use of an often unused source of auxiliary supervision: language. Inspired by recent advances in transformer-based models, we train agents with an instruction prediction loss that encourages learning temporally extended representations that operate at a high level of abstraction. Concretely, we demonstrate that instruction modeling significantly improves performance in planning environments when training with a limited number of demonstrations on the BabyAI and Crafter benchmarks. In further analysis we find that instruction modeling is most important for tasks that require complex reasoning, while understandably offering smaller gains in environments that require simple plans. More details and code can be found at https://github.com/jhejna/instruction-prediction.