Safe Behavior Specification and Planning for Autonomous Robotic Systems in Uncertain Environments

The safe operation of an autonomous robotic system is a complex endeavor, with decision-making being a pivotal element. Formal analysis of decision-making logic can be done using model checking or other formal verification approaches. However, the nondeterministic nature of realistic environments can make these approaches impractical and troublesome. Constraint-based planning approaches have been shown to be capable of generating policies for a system to reach its goals while abiding safety constraints. We extend such a constraint-based approach, Tumato, to support non-deterministic outcomes of actions. Actions have one specific intended result, yet can be modeled to have alternative outcomes that may realistically occur. The adapted Tumato solver generates a policy that enables the system to reach its goals in a safe manner even when alternative outcomes of actions occur. Furthermore, we introduce a purely declarative way of defining safety in Tumato, increasing its expressiveness and facilitating the specification of actual safe behavior. Finally, we add cost or duration values to actions, enabling the solver to restore safety when necessary in the most preferred way.