Collaborative multi-robot autonomy via distributed optimization
- The high complexity of real-world tasks necessitates the deployment of multi-robot teams, equipped with different onboard sensors and capabilities. Typically, no robot has all the capabilities required to complete the specified task; as a result, the multi-robot team must collaborate together to realize the shared objective. This thesis presents distributed algorithms that enable multi-robot teams to operate autonomously while working collaboratively. Specifically, we leverage distributed optimization to derive efficient algorithms that enable multi-robot autonomy in core robotics applications, such as autonomous transportation, search-and-rescue problems, and collaborative object manipulation. Our algorithms are fully-distributed (i.e., our algorithms do not require a central node (station) for any computation or coordination). Rather, each robot communicates with its one-hop (immediate) neighbors over a peer-to-peer communication network. We introduce a distributed optimization algorithm for problems consisting of optimization variables that can be partitioned among the robots. The proposed algorithm enables robots to avoid unnecessary optimization over irrelevant problem variables, resulting in greater computation and communication efficiency. Problems of this form are prevalent in the field of robotics, e.g., trajectory optimization. We present specialized algorithms for contact-implicit trajectory optimization and model predictive control. Our distributed contact-implicit optimization algorithm enables each robot to make and break contact, as required, to perform complex object manipulation tasks through narrow areas. Further, our distributed algorithm for model predictive control enables each robot to compute the required forces and torques needed to successfully complete a specified multi-robot task. Our algorithms eliminate the need for each robot to compute the torques of other robots, which may not be directly relevant to its performance on the given task. In addition, we present distributed algorithms for multi-robot target tracking and task assignment. Through maximum a-posteriori optimization, we derive a distributed trajectory estimation algorithm that enables each robot to estimate the trajectory of the target without sharing its local (potentially high-dimensional) observations. In addition, we introduce three algorithms for the multi-robot task assignment problem. These algorithms are designed for a broad class of multi-robot systems, considering the broad range of computation and communication resources available to multi-robot teams. In our distributed algorithms, each robot does not share its local problem data (including its local observations) and its objective and constraint functions. Rather, each robot communicates its local problem variable, which is of a generally lower-dimension, reducing the computation and communication overhead associated with our algorithms.
|Type of resource
|electronic resource; remote; computer; online resource
|1 online resource.
|Shorinwa, Olaoluwa Mosopefoluwa
|Degree committee member
|Stanford University, School of Engineering
|Stanford University, Department of Mechanical Engineering
|Statement of responsibility
|Submitted to the Department of Mechanical Engineering.
|Thesis Ph.D. Stanford University 2023.
- © 2023 by Olaoluwa Mosopefoluwa Shorinwa
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...