Coordination of Multirobot Systems Under Temporal Constraints

Abstract

Multirobot systems have great potential to change our lives by increasing efficiency or decreasing costs in many applications, ranging from warehouse logistics to construction. They can also replace humans in dangerous scenarios, for example in a nuclear disaster cleanup mission. However, teleoperating robots in these scenarios would severely limit their capabilities due to communication and reaction delays. Furthermore, ensuring that the overall behavior of the system is safe and correct for a large number of robots is challenging without a principled solution approach. Ideally, multirobot systems should be able to plan and execute autonomously. Moreover, these systems should be robust to certain external factors, such as failing robots and synchronization errors and be able to scale to large numbers, as the effectiveness of particular tasks might depend directly on these criteria. This thesis introduces methods to achieve safe and correct autonomous behavior for multirobot systems.

Firstly, we introduce a novel logic family, called counting logics, to describe the high-level behavior of multirobot systems. Counting logics capture constraints that arise naturally in many applications where the identity of the robot is not important for the task to be completed. We further introduce a notion of robust satisfaction to analyze the effects of synchronization errors on the overall behavior and provide complexity analysis for a fragment of this logic.

Secondly, we propose an optimization-based algorithm to generate a collection of robot paths to satisfy the specifications given in counting logics. We assume that the robots are perfectly synchronized and use a mixed-integer linear programming formulation to take advantage of the recent advances in this field. We show that this approach is complete under the perfect synchronization assumption. Furthermore, we propose alternative encodings that render more efficient solutions under certain conditions. We also provide numerical results that showcase the scalability of our approach, showing that it scales to hundreds of robots.

Thirdly, we relax the perfect synchronization assumption and show how to generate paths that are robust to bounded synchronization errors, without requiring run-time communication. However, the complexity of such an approach is shown to depend on the error bound, which might be limiting. To overcome this issue, we propose a hierarchical method whose complexity does not depend on this bound. We show that, under mild conditions, solutions generated by the hierarchical method can be executed safely, even if such a bound is not known.

Finally, we propose a distributed algorithm to execute multirobot paths while avoiding collisions and deadlocks that might occur due to synchronization errors. We recast this problem as a conflict resolution problem and characterize conditions under which existing solutions to the well-known drinking philosophers problem can be used to design control policies that prevents collisions and deadlocks. We further provide improvements to this naive approach to increase the amount of concurrency in the system. We demonstrate the effectiveness of our approach by comparing it to the naive approach and to the state-of-the-art.