A Safe and Robust Multi-Agent Motion Planning Framework for Urban Air Mobility

Over the past few decades, significant advancements have been made in the flight of unmanned aerial vehicles (UAVs) as well as their usage in urban air environments. This is referred to as the urban air mobility (UAM) problem, and the UAM problem has emerged as a leading challenge for these drones in recent years. Agents must be capable of real-time dynamic obstacle avoidance, planning motion in accordance with their own constraints, and detecting system failures or potential adversarial attacks. They must also be prepared to react to the behavior of other agents in the environment expressing a variety of motion planning strategies, which can range from cooperative motion planning, independent motion planning, and even planning motion around potential adversaries. These requirements may not only demand improved motion planning strategies, but may also necessitate quick development of new autonomous systems to employ these strategies. In this thesis, we consider numerous challenges that face the development of UAM and propose a urban air motion planning framework for a number of cooperating “player agents” in the environment. We formulate a model-free method of learning the optimal control for the kinodynamic motion problem, and an algorithm to predict the behavior of independent or adversarial agents in the environment through a cognitive hierarchy approach. Additionally, we use repeated observations of numerous agents to mitigate any potential noise in agent observations. This thesis also considers the challenges of modeling the effects of constraints present on a UAV system, and proposes an efficient method to learn a binary “go/no-go” classifier by selecting informative training points via a soft actor/critic (SAC) framework. We also use this same framework to augment target system inputs to ensure safe control online, as well as monitor the system to detect any potential hardware faults or adversarial attack. Both our motion planning framework and our efficient binary classification algorithm are demonstrated to be effective with numerous examples and simulations.

Level thesis:degree_level

Doctoral

Department dc:contributor.department

Aerospace Engineering

Grantor dc:publisher

Georgia Institute of Technology

Year dc:date.issued

2025

Author dc:creator

• Netter, Joshua

Advisor dc:contributor.advisor

• Vamvoudakis, Kyriakos G.

Committee members dc:contributor.committeemember

• Athanasiou, Christos
• Li, Sarah
• Coogan, Samuel
• Walsh, Timothy

• Multi-agent motion planning
• Optimal control
• Reinforcement learning

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/1853/78703

OAI identifier oai:identifier

oai:repository.gatech.edu:1853/78703