Abstract:
Multi-agent robotic teaming is the only realistic solution to many large-scale autonomous operations. Conventionally, operations are modeled as a set of tasks that are largely decoupled from each other and the environment at execution time. However, this operational model fails when the successful execution of a task requires multiple agents to synchronize their actions and adapt those actions to new operating information. This thesis addresses this limitation by explicitly encoding these required inter-agent operational couplings as execution-time synchronization constraints in a hierarchical task allocation and task execution framework. We apply our multi-agent coordination framework to robotic convoy operations to demonstrate its capability to realize solutions to operations with a high degree of synchronized multi-agent collaboration.

We then discuss how environmental interaction fits into that framework, bypassing common modeling assumptions which decouple environmental information from the task model. These assumptions cause agents to passively react to or avoid interactions with the environment, often leading to conservative agent behaviors and a false understanding of action or task feasibility. By relaxing traditional notions of obstacle avoidance in interleaved motion planning strategies, we enable an expanded feasibility of tasks that would otherwise be impossible using traditional passive avoidance behaviors. We then apply the benefits of this approach to robotic convoy operations in unstructured environments, demonstrating how an interaction-aware framework gives rise to new opportunities for synchronized multi-agent collaborations.

Finally, erroneous environmental information can have catastrophic effects on task allocation and multi-agent coordination. To mitigate these effects, we imbue our multi-agent team with the capability to communicate and adapt to newly discovered environmental information. To ensure this communication capability even in communication-deprived unstructured environments, we design a wireless ad hoc network construction technique that maintains an observed minimum signal strength between agents. With this assured communication, we can then address dynamic task allocation problems arising from inaccurate environmental data by utilizing the agents themselves as mobile environmental sensors. We then demonstrate robotic convoy operations in an unstructured environment where both agent and team reallocation and rerouting are required in response to a priori unknown information about the environment. By addressing the confluence of coupling effects, our framework effectively addresses a class of synchronized task allocation and execution problems and extends the capabilities of multi-agent robotic systems to new operational paradigms and requirements.

Thesis Committee Members:
Sebastian Scherer​ (Chair​)
Matthew Travers​ (Chair)
John Dolan
Tuomas Sandholm
Ali-akbar Agha-mohammadi​ (Field AI​)

More Information