Scalable Autonomous Vehicle Safety Validation through Dynamic Programming and Scene Decomposition

Corso, Anthony; Lee, Ritchie; Kochenderfer, Mykel J.

doi:10.1109/itsc45102.2020.9294636

Cited by 10 publications

(12 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While most importance sampling approaches focus on the entire space of disturbance trajectories, the approach of Corso, Lee, and Kochenderfer [114] uses the framework of sequential decision making to find the optimal importance sampling policy q(x | s) for a simulation state s. It is shown in that work that for a Markovian system and simulator, the optimal importance sampling policy is given by…”

Section: Approximate Dynamic Programmingmentioning

confidence: 99%

“…Local approximation dynamic programming and Monte Carlo estimations are two successful approaches for solving eq. ( 61) [114].…”

Section: Approximate Dynamic Programmingmentioning

confidence: 99%

“…The approach presented by Corso, Lee, and Kochenderfer [114] involves decomposing the simulator state and disturbance spaces into independent components, and finding failures for each component. Originally done in the context of multiple distinct actors in a simulated driving environment, this approach can be used with any simulation that has components that can be simulated individually.…”

Section: State Space Decompositionmentioning

confidence: 99%

“…If the fusion function is simple, then it may not capture the joint interactions between multiple disturbance components [114]. To address this problem, a global correction factor can be learned from the full simulation.…”

Section: State Space Decompositionmentioning

confidence: 99%

See 3 more Smart Citations

A Survey of Algorithms for Black-Box Safety Validation of Cyber-Physical Systems

Corso,

Moss,

Koren

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Autonomous and semi-autonomous systems for safety-critical applications require rigorous testing before deployment. Due to the complexity of these systems, formal verification may be impossible and real-world testing may be dangerous during development. Therefore, simulation-based techniques have been developed that treat the system under test as a black box during testing. Safety validation tasks include finding disturbances to the system that cause it to fail (falsification), finding the most-likely failure, and estimating the probability that the system fails. Motivated by the prevalence of safety-critical artificial intelligence, this work provides a survey of state-of-the-art safety validation techniques with a focus on applied algorithms and their modifications for the safety validation problem. We present and discuss algorithms in the domains of optimization, path planning, reinforcement learning, and importance sampling. Problem decomposition techniques are presented to help scale algorithms to large state spaces, and a brief overview of safety-critical applications is given, including autonomous vehicles and aircraft collision avoidance systems. Finally, we present a survey of existing academic and commercially available safety validation tools.

show abstract

Section: Approximate Dynamic Programmingmentioning

confidence: 99%

“…Local approximation dynamic programming and Monte Carlo estimations are two successful approaches for solving eq. ( 61) [114].…”

Section: Approximate Dynamic Programmingmentioning

confidence: 99%

Section: State Space Decompositionmentioning

confidence: 99%

Section: State Space Decompositionmentioning

confidence: 99%

See 2 more Smart Citations

A Survey of Algorithms for Black-Box Safety Validation of Cyber-Physical Systems

Corso,

Moss,

Koren

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Despite overall success, sampling-based approaches, such as reinforcement learning, suffer from two key drawbacks. They fail to find particularly sparse failures [9]. Additionally, they gather data from highfidelity simulators, which may exceed computational budgets [10].…”

Section: Introductionmentioning

confidence: 99%

Safety Verification of Autonomous Systems: A Multi-Fidelity Reinforcement Learning Approach

Beard¹,

Baheri²

2022

Preprint

View full text Add to dashboard Cite

As autonomous and semi-autonomous agents become more integrated with society, validation of their safety is increasingly important. The scenarios under which they are used, however, can be quite complicated; as such, formal verification may be impossible. To this end, simulationbased safety verification is being used more frequently to understand failure scenarios for the most complex problems. Recent approaches, such as adaptive stress testing (AST), use reinforcement learning, making them prone to excessive exploitation of known failures, limiting coverage of the space of failures. To overcome this, the work below defines a class of Markov decision processes, the knowledge MDP, which captures information about the learned model to reason over. More specifically, by leveraging, the "knows what it knows" (KWIK) framework, the learner estimates its knowledge (model estimates and confidence, as well as assumptions) about the underlying system. This formulation is vetted through MF-KWIK-AST which extends bidirectional learning in multiple fidelities (MF) of simulators to the safety verification problem. The knowledge MDP formulation is applied to detect convergence of the model, penalizing this behavior to encourage further exploration. Results are evaluated in a grid world, training an adversary to intercept a system under test. Monte Carlo trials compare the relative sample efficiency of MF-KWIK-AST to learning with a single-fidelity simulator, as well as demonstrate the utility of incorporating knowledge about learned models into the decision making process.

show abstract

Intersection Focused Situation Coverage-Based Verification and Validation Framework for Autonomous Vehicles Implemented in CARLA

Tahir

Alexander

2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Autonomous Vehicles (AVs) i.e., self-driving cars, operate in a safetycritical domain, since errors in the autonomous driving software can lead to huge losses. Statistically, road intersections which are a part of the AVs operational design domain (ODD), have some of the highest accident rates. Hence, testing AVs to the limits on road intersections and assuring their safety on road intersections is pertinent, and thus the focus of this paper. We present a situation coverage-based (SitCov) AV-testing framework for the verification and validation (V&V) and safety assurance of AVs, developed in an open-source AV simulator named CARLA. The SitCov AV-testing framework focuses on vehicle-to-vehicle (V2V) interaction on a road intersection under different environmental conditions and intersection configuration situations (start/goal locations), using situation coverage criteria for automatic test suite generation for safety assurance of AVs. We have developed an ontology for intersection situations, and used it to generate a situation hyperspace i.e., the space of all possible situations arising from that ontology. For the evaluation of our SitCov AV-testing framework, we have seeded multiple faults in our ego AV, and compared situation coveragebased and random situation generation. We have found that both generation methodologies trigger around the same number of seeded faults, but the situation coverage-based generation tells us a lot more about the weaknesses of the autonomous driving algorithm of our ego AV, especially in edge-cases. Our code is publicly available online and since the simulation software (CARLA) is opensource, anyone can use our SitCov AV-testing framework and use it or build further on top of it. This paper aims to contribute to the domain of V&V and development of AVs, not only from a theoretical point of view, but also from the viewpoint of an open-source software contribution and releasing a flexible/effective tool for V&V and development of AVs.

show abstract

Scalable Autonomous Vehicle Safety Validation through Dynamic Programming and Scene Decomposition

Cited by 10 publications

References 20 publications

A Survey of Algorithms for Black-Box Safety Validation of Cyber-Physical Systems

A Survey of Algorithms for Black-Box Safety Validation of Cyber-Physical Systems

Safety Verification of Autonomous Systems: A Multi-Fidelity Reinforcement Learning Approach

Intersection Focused Situation Coverage-Based Verification and Validation Framework for Autonomous Vehicles Implemented in CARLA

Contact Info

Product

Resources

About