Explainable Robotic Plan Execution Monitoring Under Partial Observability

Abstract

Successful plan generation for autonomous systems is necessary but not sufficient to guarantee reaching a goal state by an execution of a plan. Various discrepancies between an expected state and the observed state may occur during the plan execution (e.g., due to failure of robot parts) and these discrepancies may lead to plan failures. For that reason, autonomous systems should be equipped with execution monitoring algorithms so that they can autonomously recover from such discrepancies.We introduce a plan execution monitoring algorithm that operates under partial observability. This algorithm relies on novel formal methods for hybrid prediction, diagnosis and explanation generation, and planning. The prediction module generates an expected state after the execution of a part of the plan from an incomplete state, to check for discrepancies. The diagnostic reasoning module generates meaningful hypotheses to explain failures of robot parts. Unlike the existing diagnosis methods, the previous hypotheses can be revised, based on new partial observations, increasing the accuracy of explanations as further information becomes available. The replanning module considers these explanations while computing a new plan that would avoid such failures. All these reasoning modules are hybrid in that they combine high-level logical reasoning with low-level feasibility checks based on probabilistic methods. We experimentally show that these hybrid reasoning modules improve the performance of plan execution monitoring in service robotics applications with multiple bimanual mobile robots.To evaluate the performance and to understand the applicability of the proposed execution monitoring algorithm, we introduce an execution simulation algorithm. This algorithm is based on a formal method that allows generation of

Failure is an option: Task and Motion Planning with Failing Executions

“…But a simulator might not be a perfect model of the real world, presenting challenges in applying the policies learned in simulation to real robots, known as the sim-to-real gap [46]. Some methods address this problem, but mostly only focus on recovery from failures caused by unmodeled dynamics, physical interactions, or system errors [13], [29].…”

“…Plan execution monitoring methods [27] specifically focus on detecting the difference between world states and expected states, and recovering from the discrepancies. There have been works that consider execution failures by diagnosing broken parts of the robotic system with either fully observable [28] or partially observable state spaces [29]. Another work reacts to temporal or resource constraints that emerge at execution time [30].…”

A Taxonomy of Explanation Types and Need Indicators in Human–Agent Collaborations

“…Self-monitoring explanations shed light on a robot's comprehension of its own state, that is, the state of software and hardware modules, including potential malfunctioning [49,50]. An example of an explanation-demanding question is "Why is your battery so low?"…”