University of Twente Student Theses


Autonomous mapping and navigation of an unknown environment using a reinforcement learning approach

Schulte, Rob (2020) Autonomous mapping and navigation of an unknown environment using a reinforcement learning approach.

[img] PDF
Abstract:The capability of autonomous (mobile) robots to navigate their environment is a critical first step in the development towards the deployment of mobile robots in our daily lives. Traditional path planners utilize a map of the environment in order to plan a collision-free path. However, a-priori map knowledge is not always precise or, depending on the task, not available at all. Other approaches which construct a map of the environment during the navigation process, such as (active) SLAM approaches, require pre-coded navigational directives or path planners in order to obtain a collision-free path. In this work, we formulate the navigation and mapping task in a-priori unknown environments as a reinforcement learning problem such that the previously mentioned problems can be alleviated. In particular, a DDPG and SLAM approach is used in order to navigate and map known and unknown environments using sensory information obtained. The main novelty of this work is aiding the robots navigational process by combining SLAM information and a concept of reachability in the state-space within the reward function definition. In this work, four different reward functions are compared to several domestic simulation environments with different sizes and complexities. The used reward structures are evaluated in the training environment in addition to a-priori unknown environments to test the generalization capabilities. The evaluation indicates that the proposed curiosity approach was able to cut down training time significantly within the used training environments. Other used performance metrics such as trajectories and success ratio also signify that the curiosity approach surpasses all other approaches in both the training and testing environments.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Electrical Engineering MSc (60353)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page