Georgia Institute of Technology
Model-free Multi-Agent Motion Planning with Non-Uniform Sensors
Abstract
dc:description.abstractAutonomous robotic systems are becoming an increasingly viable solution to many of the world's complex problems. For a system to operate autonomously, it must understand what task needs to be done, create a plan for how to perform that task based on sensor input, and finally, execute that plan through its actuators. Often a stationary robotic system is not enough for the tasks we want robots to do for us, therefore we turn to mobile robotic systems to navigate through an environment to reach some goal location of a particular task. This thesis focuses on planning a viable path for multiple robots through a shared, unknown environment and the coordinated execution of that plan as the environment changes. Furthermore, we restrict the problem such that not all robots have the ability to sense their environment. These "blind" robots will rely on other robots to safely guide them through the environment. By combining aspects of reinforcement learning with sampling-based path planning methods, we can develop a multi-agent motion planning framework to guide all robots safely through the environment. Reinforcement learning provides the advantage of being able to learn how to control a system without needing to know the system's dynamics. This process is done by an actor/critic based Q-learning framework that learns by exploring different inputs to converge towards stable control. Additionally, a sampling-based path planner, designed for dynamic environments, is used in the coupled state space to find a path for all robots to safely navigate the environment. The unknown system dynamics and the blind robots will require extra constraints to be placed on the path planning. These constraints will be implemented through obstacle augmentation and the introduction of sensor space. Simulations for the proposed framework are developed for both 2D and 3D environments to verify the approach used.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Staatz, Dylan Matthew
- Advisor dc:contributor.advisor
-
- Vamvoudakis, Kyriakos G.
- Committee members dc:contributor.committeemember
-
- Wardi, Yorai Y.
- Coogan, Samuel
Subjects
dc:subject × 3Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- http://hdl.handle.net/1853/66545