Safety Engineering for Autonomous Vehicles

Adler, Rasmus; Feth, Patrik; Schneider, Daniel

doi:10.1109/dsn-w.2016.30

Cited by 21 publications

(9 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…If the controller issues commands that breach the safety cages, the safety cages step in and attempt to recover the system back to a safe state. This type of approach has been used to guarantee the safety of complex controllers in different domains such as robotics [ 44 , 45 , 46 , 47 ], aerospace [ 48 ], and automotive applications [ 49 , 50 , 51 ]. Heckemann et al [ 18 ] suggested that these safety cages could be used to ensure the safety of black box systems in autonomous vehicles by utilising the vehicle’s sensors to monitor the state of the environment, and then limiting the actions of the vehicle in safety-critical scenarios.…”

Section: Related Workmentioning

confidence: 99%

“…Heckemann et al [ 18 ] suggested that these safety cages could be used to ensure the safety of black box systems in autonomous vehicles by utilising the vehicle’s sensors to monitor the state of the environment, and then limiting the actions of the vehicle in safety-critical scenarios. Demonstrating the effectiveness of this approach, Adler et al [ 49 ] proposed five safety cages based on the Automotive Safety Integrity Levels (ASIL) defined by ISO26262 [ 52 ] to improve the safety of an autonomous vehicle with machine learning based controllers. Focusing on path planning in urban environments, Yurtsever et al [ 53 ] combined RL with rule-based path planning to provide safety guarantees in autonomous driving.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Weakly Supervised Reinforcement Learning for Autonomous Highway Driving via Virtual Safety Cages

Kuutti

Bowden

Fallah

2021

Sensors

View full text Add to dashboard Cite

The use of neural networks and reinforcement learning has become increasingly popular in autonomous vehicle control. However, the opaqueness of the resulting control policies presents a significant barrier to deploying neural network-based control in autonomous vehicles. In this paper, we present a reinforcement learning based approach to autonomous vehicle longitudinal control, where the rule-based safety cages provide enhanced safety for the vehicle as well as weak supervision to the reinforcement learning agent. By guiding the agent to meaningful states and actions, this weak supervision improves the convergence during training and enhances the safety of the final trained policy. This rule-based supervisory controller has the further advantage of being fully interpretable, thereby enabling traditional validation and verification approaches to ensure the safety of the vehicle. We compare models with and without safety cages, as well as models with optimal and constrained model parameters, and show that the weak supervision consistently improves the safety of exploration, speed of convergence, and model performance. Additionally, we show that when the model parameters are constrained or sub-optimal, the safety cages can enable a model to learn a safe driving policy even when the model could not be trained to drive through reinforcement learning alone.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Weakly Supervised Reinforcement Learning for Autonomous Highway Driving via Virtual Safety Cages

Kuutti

Bowden

Fallah

2021

Sensors

View full text Add to dashboard Cite

show abstract

“…At the operational level, fail-over behavior can be activated when faults in the field are discovered. Fail-over behaviors can bring a system into a safe state in face of hazardous events [37].…”

Section: Platform For Runtime Evaluation Of Goalsmentioning

confidence: 99%

Goals within Trust-based Digital Ecosystems

Cioroaica

Purohit

Bühnová

et al. 2021

2021 IEEE/ACM Joint 9th International Workshop on Software Engineering for Systems-of-Systems and 15th Workshop on Distributed

Self Cite

View full text Add to dashboard Cite

Within a digital ecosystem, systems and actors form coalitions for achieving common and individual goals. In a constant motion of collaborative and competitive forces and faced with the risk of malicious attacks, ecosystem participants require strong guarantees of their collaborators' trustworthiness. Evidence of trustworthy behavior derived from runtime executions can provide these trust guarantees, given that clear definition and delimitation of trust concerns exist. Without them, a base for negotiating expectations, quantifying achievements and identifying strategical attacks cannot be established and attainment of strategic benefits relies solely on vulnerable collaborations. In this paper we examine the relationship between goals and trust and we introduce a formalism for goal representation. We delimit the trust concerns with anti-goals. The anti-goals set the boundaries within which we structure the trust analysis and build up evidence for motivated attacks.

show abstract

“…For autonomous vehicles, Heckemann et al [7] suggested that these techniques could be useful for ensuring the safety of complex and adaptive machine learning systems in autonomous vehicles. Adler et al [1] proposed safety cages based on five constraints such as "accelerating if a slower vehicle is closely in front" to meet the five Automotive Safety Integrity Levels (ASIL) defined in ISO26262 [8].…”

Section: Introductionmentioning

confidence: 99%

Intelligent Data Engineering and Automated Learning – IDEAL 2019

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Deep learning is a promising class of techniques for controlling an autonomous vehicle. However, functional safety validation is seen as a critical issue for these systems due to the lack of transparency in deep neural networks and the safety-critical nature of autonomous vehicles. The black box nature of deep neural networks limits the effectiveness of traditional verification and validation methods. In this paper, we propose two software safety cages, which aim to limit the control action of the neural network to a safe operational envelope. The safety cages impose limits on the control action during critical scenarios, which if breached, change the control action to a more conservative value. This has the benefit that the behaviour of the safety cages is interpretable, and therefore traditional functional safety validation techniques can be applied. The work here presents a deep neural network trained for longitudinal vehicle control, with safety cages designed to prevent forward collisions. Simulated testing in critical scenarios shows the effectiveness of the safety cages in preventing forward collisions whilst under normal highway driving unnecessary interruptions are eliminated, and the deep learning control policy is able to perform unhindered. Interventions by the safety cages are also used to retrain the network, resulting in a more robust control policy.

show abstract

Safety Engineering for Autonomous Vehicles

Cited by 21 publications

References 2 publications

Weakly Supervised Reinforcement Learning for Autonomous Highway Driving via Virtual Safety Cages

Weakly Supervised Reinforcement Learning for Autonomous Highway Driving via Virtual Safety Cages

Goals within Trust-based Digital Ecosystems

Intelligent Data Engineering and Automated Learning – IDEAL 2019

Contact Info

Product

Resources

About