Auxiliary Tasks Speed Up Learning PointGoal Navigation

TL;DR

This paper introduces an efficient learning method for PointGoal Navigation that leverages self-supervised auxiliary tasks and attention mechanisms, significantly reducing training time and improving performance over previous approaches.

Contribution

It proposes a novel combination of auxiliary tasks with attention to enhance sample efficiency and speed up PointNav learning, achieving state-of-the-art results with fewer frames.

Findings

5.5x faster training to reach previous SOTA performance

Improved SPL by 0.16 at 40M frames

Auxiliary tasks combined with attention outperform naive methods

Abstract

PointGoal Navigation is an embodied task that requires agents to navigate to a specified point in an unseen environment. Wijmans et al. showed that this task is solvable but their method is computationally prohibitive, requiring 2.5 billion frames and 180 GPU-days. In this work, we develop a method to significantly increase sample and time efficiency in learning PointNav using self-supervised auxiliary tasks (e.g. predicting the action taken between two egocentric observations, predicting the distance between two observations from a trajectory,etc.).We find that naively combining multiple auxiliary tasks improves sample efficiency,but only provides marginal gains beyond a point. To overcome this, we use attention to combine representations learnt from individual auxiliary tasks. Our best agent is 5.5x faster to reach the performance of the previous state-of-the-art, DD-PPO, at 40M frames, and improves on DD-PPO's performance at 40M frames by 0.16 SPL. Our code is publicly available at https://github.com/joel99/habitat-pointnav-aux.