Formal synthesis of control and communication schemes
In traditional motion planning, the problem is simply specified as "go from A to B while avoiding obstacles", where A and B are two configurations or regions of interest in the robot workspace. However, a large number of robotic applications require more expressive specification languages, which allow for logical and temporal statements about the satisfaction of properties of interest. Examples include "visit A and B infinitely often, always avoid C, and do not visit D unless E vas visited before". Such task specifications cannot be trivially converted to a sequence of "go from A to B" primitives. This thesis establishes theoretical and computational frameworks for automatic synthesis of robot control and communication schemes that are correct-by-construction from task specifications given in expressive languages. We consider a purely discrete scenario, in which the dynamics of each robot is modeled as a finite discrete system. The first problem addressed in this thesis is the generation of provably-correct individual control and communication strategies for a team of robots from rich task specifications in the case when the workspace is static. The second problem relaxes this assumption and considers a scenario in which the environment changes according to some unknown patterns. It proposed a combined learning and formal synthesis approach to generate correct control policies. To tackle the first problem, we draw inspirations from the research fields of formal verification and synthesis, distributed formal synthesis, and concurrency theory. We consider a team of robots that can move among the regions of a partitioned environment and have known capabilities of servicing a set of requests that can occur in the regions of the partition. Some of these requests can be serviced by a robot individually, while some require the cooperation of groups of robots. We propose a top-down approach, in which global specifications given as Regular Expressions (RE) or Linear Temporal Logics (LTL) can be decomposed into local (individual) specifications, which can then be used to automatically synthesize robot control and communication strategies. To address the second problem, we bring together automata learning methods from the field of theoretical linguistics and techniques from temporal logic games and probabilistic model checking, to develop a provably-correct control strategy for robots moving in an environment with unknown dynamics. The robots are required to achieve a surveillance mission, in which a certain request needs to be serviced repeatedly, while the expected time in between consecutive services is minimized and additional temporal logic constraints are satisfied. We define a fragment of Linear Temporal Logic (LTL) to describe such a mission. We consider a single agent case at first and then extend the results to multi-agent systems. To this end, we apply approximate dynamic programming to our computational framework, which leads to significant reduction of computational time. To demonstrate the proposed theoretical and computational frameworks, we implement the derived algorithms in two experimental platforms, the Robotic Urban-Like Environment (RULE) and the Robotic InDoor-like Environment (RIDE). We assign tasks to the team using Regular Expressions or Linear Temporal Logics over requests occurring at regions in the environment. The robots are automatically deployed to complete the missions.
Thesis (Ph.D.)--Boston University