Thesis title: Exploiting Two-layer Redundancy: Control and Task Allocation for Heterogeneous Multi-Agent Systems
In multirobot multitask scenarios, redundancy is a fundamental feature arising both from the presence of more agents than tasks and from surplus input resources relative to the controlled variables. This redundancy gives rise to two interrelated problems. First, determining which agent should execute which task to collectively achieve a global objective defines the task allocation problem. This requires accounting for the heterogeneous nature of agents and the constraints associated with task execution. Second, input redundancy enables the control allocation problem, which seeks to exploit extra degrees of freedom to fulfill secondary objectives, such as minimizing power consumption or enhancing fault tolerance. In practice, however, control allocation must account for input constraints that may give rise to the windup problem.
This thesis develops a comprehensive framework to simultaneously address task and control allocation in heterogeneous redundant multi-agent systems. The proposed cascaded architecture integrates input redundancy within the task allocation process, enabling the exploitation of surplus degrees of freedom to mitigate windup effects while executing tasks collaboratively.
The first part introduces a decentralized dynamic control allocation approach for weakly redundant multi-agent systems. This method is based on a dynamic output invisible allocator composed of an optimizer and an annihilator, designed to overcome saturation conditions and recover intended behavior. The allocation policy relies solely on local information, ensuring scalability.
The second part develops an optimization-based task allocation framework for heterogeneous redundant multi-agent systems. This framework leverages input redundancy information to assign tasks to the most capable and redundant agents, enabling cooperative task execution. The allocation problem is formulated as a Mixed-Integer Linear Program (MILP), enabling the systematic handling of agent heterogeneity and execution constraints. This leads to a fully decentralized control and task allocation framework for heterogeneous redundant systems.
The final part presents two independent applications of the proposed task allocation framework. First, task allocation is applied to autonomous collaborative ship collision avoidance, integrating maritime traffic regulations within the allocation policy. Second, task allocation is applied to Earth observation operations in multi-satellite systems, extending the formulation to support secondary objectives such as maximizing coverage of areas of interest and minimizing data latency to ground stations. These case studies demonstrate the framework’s capability to handle real-world complexities, such as dynamic operational constraints in maritime navigation and communication limitations in multi-satellite coordination.