Recent accidents involving autonomous vehicles prompt us to consider how we can engineer an autonomous vehicle which always obeys traffic rules. This is particularly challenging because traffic rules are rarely specified at the level of detail an engineer would expect. Hence, it is nearly impossible to formally monitor behaviours of autonomous vehicles-which are expressed in terms of position, velocity, and acceleration-with respect to the traffic rules-which are expressed by vague concepts such as "maintaining safe distance". We show how we can use the Isabelle theorem prover to do this by first codifying the traffic rules abstractly and then subsequently concretising each atomic proposition in a verified manner. Thanks to Isabelle's code generation, we can generate code which we can use to monitor the compliance of traffic rules formally.
Abstract-One significant barrier in introducing autonomous driving is the liability issue of a collision; e.g. when two autonomous vehicles collide, it is unclear which vehicle should be held accountable. To solve this issue, we view traffic rules from legal texts as requirements for autonomous vehicles. If we can prove that an autonomous vehicle always satisfies these requirements during its operation, then it cannot be held responsible in a collision. We present our approach by formalising a subset of traffic rules from the Vienna Convention on Road Traffic for highway scenarios in Isabelle/HOL.
Abstract. One barrier in introducing autonomous vehicle technology is the liability issue when these vehicles are involved in an accident. To overcome this, autonomous vehicle manufacturers should ensure that their vehicles always comply with traffic rules. This paper focusses on the safe distance traffic rule from the Vienna Convention on Road Traffic. Ensuring autonomous vehicles to comply with this safe distance rule is problematic because the Vienna Convention does not clearly define how large a safe distance is. We provide a formally proved prescriptive definition of how large this safe distance must be, and correct checkers for the compliance of this traffic rule. The prescriptive definition is obtained by: 1) identifying all possible relative positions of stopping (braking) distances; 2) selecting those positions from which a collision freedom can be deduced; and 3) reformulating these relative positions such that lower bounds of the safe distance can be obtained. These lower bounds are then the prescriptive definition of the safe distance, and we combine them into a checker which we prove to be sound and complete. Not only does our work serve as a specification for autonomous vehicle manufacturers, but it could also be used to determine who is liable in court cases and for online verification of autonomous vehicles' trajectory planner.
Abstract-While a number of efficient methods have been proposed for approximating backward reachable sets, no synthesis method via backward reachable sets has been developed for estimating and enlarging the region of attraction (RA). This paper shows how to use backward reachable sets to enlarge the estimate of the RA of linear discrete-time systems, by using an optimal static feedback controller. Two controller design methods are provided: the first method enlarges the estimate of the RA via invariant sets, whose existence is ensured by zonotope containment; the second method provides the optimal control input by using Lyapunov stability and quadratic stabilization. The backward reachable set is represented by zonotopes which give a good compromise between accuracy and efficiency. The effectiveness of both methods is illustrated by a numerical example.
Autonomous vehicles are safety-critical cyber-physical systems. To ensure their correctness, we use a proof assistant to prove safety properties deductively. This paper presents a formally verified motion planner based on manoeuvre automata in Isabelle/HOL. Two general properties which we ensure are numerical soundness (the absence of floating-point errors) and logical correctness (satisfying a plan specified in linear temporal logic). From these two properties, we obtain a motion planner whose correctness only depends on the validity of the models of the ego vehicle and its environment.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.