Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/209013
Title: LEARNING TO NAVIGATE WITHOUT A MAP
Authors: ZHANG WEI
Keywords: Robot Navigation, Motion Planning, Deep Reinforcement Learning, Reactive and Sensor-Based Planning, Collision Avoidance, Mapless Navigation.
Issue Date: 15-Apr-2021
Citation: ZHANG WEI (2021-04-15). LEARNING TO NAVIGATE WITHOUT A MAP. ScholarBank@NUS Repository.
Abstract: Self-navigation, allowing mobile robots to reach their destinations without collisions, is a fundamental skill required for mobile robots. Recently, deep reinforcement learning (DRL) is employed to address the mapless navigation problem and achieves notable successes. However, the trained controller suffers from great performance degradation when deployed to real robots for real-world navigation tasks. In this thesis, four approaches are proposed to address these challenges. First, an efficient but simple pre-processing approach is proposed to accelerate training and enhance the agent’s generalization capability. Second, a new DRL model is proposed to address range data obtained from different range sensors with varied installation positions. Third, the MSL-DVST approach is proposed to adaptively transfer the DRL-based navigation skill of the meta-agent to robots with varied dimensions. Last, a hierarchical control framework is proposed to enhance the long-term planning capability of the robot trained with DRL.
URI: https://scholarbank.nus.edu.sg/handle/10635/209013
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
ZhangW.pdf8.08 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

45
checked on Oct 6, 2022

Download(s)

5
checked on Oct 6, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.