End-To-End Deep Reinforcement Learning for First-Person Pedestrian Visual Navigation in Urban Environments
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
Standard
IEEE-RAS International Conference on Humanoid Robots. 2022. (IEEE-RAS International Conference on Humanoid Robots).
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
Harvard
APA
Vancouver
Author
Bibtex - Download
}
RIS (suitable for import to EndNote) - Download
TY - GEN
T1 - End-To-End Deep Reinforcement Learning for First-Person Pedestrian Visual Navigation in Urban Environments
AU - Xue, Honghu
AU - Song, Rui
AU - Petzold, Julian
AU - Hein, Benedikt
AU - Hamann, Heiko
AU - Rueckert, Elmar
PY - 2022/9/26
Y1 - 2022/9/26
N2 - We solve a visual navigation problem in an urban setting via deep reinforcement learning in an end-to-end manner. A major challenge of a first-person visual navigation problem lies in severe partial observability and sparse positive experiences of reaching the goal. To address partial observability, we propose a novel 3D-temporal convolutional network to encode sequential historical visual observations, its effectiveness is verified by comparing to a commonly-used frame-stacking approach. For sparse positive samples, we propose an improved automatic curriculum learning algorithm NavACL+, which proposes meaningful curricula starting from easy tasks and gradually generalizes to challenging ones. NavACL+ is shown to facilitate the learning process, greatly improve the task success rate on difficult tasks by at least 40% and offer enhanced generalization to different initial poses compared to training from a fixed initial pose and the original NavACL algorithm.
AB - We solve a visual navigation problem in an urban setting via deep reinforcement learning in an end-to-end manner. A major challenge of a first-person visual navigation problem lies in severe partial observability and sparse positive experiences of reaching the goal. To address partial observability, we propose a novel 3D-temporal convolutional network to encode sequential historical visual observations, its effectiveness is verified by comparing to a commonly-used frame-stacking approach. For sparse positive samples, we propose an improved automatic curriculum learning algorithm NavACL+, which proposes meaningful curricula starting from easy tasks and gradually generalizes to challenging ones. NavACL+ is shown to facilitate the learning process, greatly improve the task success rate on difficult tasks by at least 40% and offer enhanced generalization to different initial poses compared to training from a fixed initial pose and the original NavACL algorithm.
KW - Autonomous Navigation
KW - Deep Learning
KW - mobile navigation
M3 - Conference contribution
T3 - IEEE-RAS International Conference on Humanoid Robots
BT - IEEE-RAS International Conference on Humanoid Robots
ER -
