Performance verification for robot missions in uncertain environments

Abstract

h i g h l i g h t s• This paper addresses the challenges involved in building a software tool for automatically verifying the behavior of multi-robot waypoint missions using formal methods.• Missions can include uncertainly located obstacles and uncertain environment geometry as well as uncertainty in robot motion.• We leverage a unique approach, VIPARS, to verifying performance guarantees for autonomous behavior-based robot software based on a combination of static analysis and Bayesian networks.• Two approaches to modeling probabilistic localization for verification are presented: a high-level approach and an approach that allows run-time localization code to be embedded within verification.• Verification and experimental validation results are presented for several autonomous robot missions, demonstrating the accuracy of verification and the mission-specific benefit of localization. a b s t r a c tEstablishing a-priori mission performance guarantees is crucial if autonomous robots are to be used with confidence in missions where failure could incur high costs in life and property damage. Automatic mission software verification, in addition to simulation and experimental benchmarking, is a key component of the solution for establishing performance guarantees. This component requires automatically verifying that the software constructed by the mission designer when executed in a partially known environment will adhere to the performance guarantee. In prior work we developed VIPARS, a unique approach to verifying performance guarantees for autonomous behavior-based robot software based on a combination of static analysis and Bayesian networks. While that approach produced fast and accurate verification of single robot missions with robot motion uncertainty, it did not address multiple-robot missions or any form of uncertainty related to environment geometry. This paper addresses the challenges involved in building a software tool for verifying the behavior of a multi-robot waypoint mission that includes uncertainly located obstacles and uncertain environment geometry as well as uncertainty in robot motion. An approach is presented to the problem of a-priori specification of uncertain environments for robot program verification. Two approaches to modeling probabilistic localization for verification are presented: a high-level approach and an approach that allows run-time localization code to be embedded in verification. Verification and experimental validation results are presented for several autonomous robot missions, demonstrating the accuracy of verification and the mission-specific benefit of localization

Lightweight Multilingual Software Analysis

“…This is as compared to dynamic analysis, which annotates code and evaluates execution traces for single runs or statistics for sets of runs. However, in prior work [24] [25], static analysis techniques have been extended with Dynamic Bayesian Networks [26] to produce probabilities associated with variable values. Integrating this with the RDA (or a points-to) filter in MLSA would mean that the results generated, while still sound, would be less conservative and more reflective of real run-time operation.…”

Robot Mission Planning using Co-evolutionary Optimization

“…While many papers of the domain do not account for optimality, the work of Svorenova et al 9 modelled preferences and rewards so as to always focus the search towards the states satisfying the goal criterion, while also maximizing for the rewards. Lyons et al 10 further introduced uncertainties and made a framework for mission verification with multiple robots. While the current results in this cluster are very encouraging, the approaches can never scale up to the level of future demands due to the exponential complexity involved.…”

“…Let Z (t) denote the sub-task that will be performed at order sequence t, a mapping from O to Z is given by Eqs. (9) and (10).…”

Using Taint Analysis and Reinforcement Learning (TARL) to Repair Autonomous Robot Software

“…Not many software developers have the skill set necessary to formally establish program correctness according to a specification. But even if they did, formal verification of software [2] [3] and autonomous systems [4], as well as 'correct by construction' formal synthesis methods [5] all suffer from the problem that they require a designer-specified model of the environment. And that model can be incomplete because of designer oversight or oversimplification of a complex and dynamic environment.…”