The Surprising Effectiveness of Visual Odometry Techniques for Embodied PointGoal Navigation

TL;DR

This paper demonstrates that visual odometry significantly enhances embodied PointGoal navigation in realistic noisy environments, boosting success rates and execution speed without relying on GPS or Compass sensors.

Contribution

The study shows that integrating visual odometry into navigation policies improves success and speed in realistic PointGoal navigation tasks, challenging prior assumptions about sensor requirements.

Findings

Success rate increased from 64.5% to 71.7%.

Navigation speed improved by 6.4 times.

Visual odometry is highly effective under realistic noise conditions.

Abstract

It is fundamental for personal robots to reliably navigate to a specified goal. To study this task, PointGoal navigation has been introduced in simulated Embodied AI environments. Recent advances solve this PointGoal navigation task with near-perfect accuracy (99.6% success) in photo-realistically simulated environments, assuming noiseless egocentric vision, noiseless actuation, and most importantly, perfect localization. However, under realistic noise models for visual sensors and actuation, and without access to a "GPS and Compass sensor," the 99.6%-success agents for PointGoal navigation only succeed with 0.3%. In this work, we demonstrate the surprising effectiveness of visual odometry for the task of PointGoal navigation in this realistic setting, i.e., with realistic noise models for perception and actuation and without access to GPS and Compass sensors. We show that integrating visual odometry techniques into navigation policies improves the state-of-the-art on the popular Habitat PointNav benchmark by a large margin, improving success from 64.5% to 71.7% while executing 6.4 times faster.