TY - GEN
T1 - The Surprising Effectiveness of Visual Odometry Techniques for Embodied PointGoal Navigation
AU - Zhao, Xiaoming
AU - Agrawal, Harsh
AU - Batra, Dhruv
AU - Schwing, Alexander
N1 - Funding Information:
To conclude, we find classical visual odometry techniques to be surprisingly effective and yield a very strong baseline for Embodied PointGoal Navigation in a realistic setting (noisy actuation and perception; no localization sensor). Acknowledgements: Work supported in part by NSF grants #1718221, 2008387, 2045586, MRI #1725729, and NIFA 2020-67021-32799, UIUC, Samsung, Amazon and Cisco Systems Inc. (award 1377144 - thanks for access to Arcetri).
Publisher Copyright:
© 2021 IEEE
PY - 2021
Y1 - 2021
N2 - 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.
AB - 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.
UR - http://www.scopus.com/inward/record.url?scp=85121021078&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85121021078&partnerID=8YFLogxK
U2 - 10.1109/ICCV48922.2021.01582
DO - 10.1109/ICCV48922.2021.01582
M3 - Conference contribution
AN - SCOPUS:85121021078
T3 - Proceedings of the IEEE International Conference on Computer Vision
SP - 16107
EP - 16116
BT - Proceedings - 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 18th IEEE/CVF International Conference on Computer Vision, ICCV 2021
Y2 - 11 October 2021 through 17 October 2021
ER -