Safe and Robust Robot Behavior Planning via Constraint Programming

TL;DR

This paper extends a constraint-based planning approach to handle non-deterministic action outcomes, enhancing safety guarantees and expressiveness in autonomous system decision-making under uncertainty.

Contribution

It introduces non-deterministic action modeling, a declarative safety specification, and cost-aware safety restoration in Tumato, improving robustness and flexibility.

Findings

Successfully models alternative action outcomes.

Ensures safe goal achievement despite uncertainties.

Increases planning expressiveness with declarative safety.

Abstract

The safe operation of an autonomous system is a complex endeavor, one pivotal element being its decision-making. Decision-making logic can formally be analyzed using model checking or other formal verification approaches. Yet, the non-deterministic nature of realistic environments makes these approaches rather troublesome and often impractical. Constraint-based planning approaches such as Tumato have been shown to be capable of generating policies for a system to reach a stated goal and abiding safety constraints, with guarantees of soundness and completeness by construction. However, uncertain outcomes of actions in the environment are not explicitly modeled or accounted for, severely limiting the expressiveness of Tumato. In this work, we extend Tumato with support for non-deterministic outcomes of actions. Actions have a specific intended result yet can be modeled to have alternative outcomes that may realistically occur. The adapted solver generates a policy that enables reaching the 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. Finally, the addition of cost or duration values to actions enables the solver to restore safety when necessary, in the most preferred way.