A Conceptual Control System Description of Cooperative and Automated Driving in Mixed Urban Traffic With Meaningful Human Control for Design and Evaluation

Abstract

The introduction of automated vehicles means that some or all operational control over these vehicles is diverted away from a human driver to a technological system. The concept of Meaningful Human Control (MHC) was derived to address control issues over automated systems, allowing a system to explicitly consider human intentions and reasons. Applying MHC to technological systems, such as automated driving is a real challenge, and the main focus of this paper. An approach with mathematical elaboration has been developed that offers a first quantifiable operationalisation of MHC for the traffic domain and for use with automated vehicles. A major contribution lies in the taxonomification of control for MHC in the broader traffic environment, including consideration of the driver, the vehicle, the traffic environment, considering behaviour, moral standards and societal values, which are considered in a case study. The demonstration case shows the validity of the developed approach for an automated vehicle overtaking a cyclist on an urban street. This paper is one of the first to operationalise MHC to such a level of detail and opens the door to further development of the concept for technological implementation.

Meaningful human control: actionable properties for AI system development

“…The tracking and tracing conditions take the concept of meaningful human control one step closer to support practical design and development because they provide highlevel design requirements for a human-AI system to be under meaningful human control. Building on this conception, researchers developed frameworks to analyze and quantify factors affecting meaningful human control for automated vehicles [12,41,42]. However, a description of general system-level properties that could support operationalization of tracking and tracing conditions in diverse contexts is yet to be specified.…”

Realising Meaningful Human Control Over Automated Driving Systems: A Multidisciplinary Approach

“…To be able to achieve this, a taxonomy of the key core components of the systems must be constructed. Based on state-of-the-art literature from many sources, including (Amditis et al, 2012 ; Chandrasiri et al ( 2016 ); Farah et al, 2018 ; Ibañez-Guzman et al, 2012 ; Körber et al, 2015 ; McKnight & Adams, 1970 ; Sanchez et al, 2016 ; Theologus & Fleishman ( 1971 )), Calvert et al ( 2020a , 2020b ) constructed such a taxonomy identifying the core components for the Vehicle, Driver, Infrastructure and Environment systems An excerpt of the taxonomy for the Vehicle category is shown below in Fig. 3 .…”

“…For quantitative approaches, (Calvert & Mecacci, 2020 ) start with the aforementioned taxonomy of components. The two main conditions (tracking and tracing) are conceptually translated from philosophical and ethics descriptions into a framework of solid technical and cognitive connections by making use of the components.…”

“…To describe such a variable is a massive undertaking in itself and is therefore not feasible in the short term for a demonstration of MHC operationalisation. The eventual extent of operationalisation into mathematical formulations was limited in Calvert and Mecacci ( 2020 ) to the variables that allowed a specific use case to be considered. This use case focussed on a highly automated vehicle on an urban road that gradually learns how to overtake a cyclist while adhering and optimising towards human reasons that are formulated in terms of tracking and tracing with a focus on traffic safety and a timely arrival at the vehicle’s destination.…”

A practitioner’s View of Driver Training for automated driving From Driving Examiners: A Focus group Discussion

“…How these issues can help with ensuring automated driving systems to allow a meaningful form of human control, can be explained by emphasizing the two main conditions surrounding the concept of meaningful human control, namely tracking and tracing. Where tracking involves both agents and reasons [42], [43] and entails a system that always tracks its users' (proximal to distal) reasons, ranging from moral values to steering actions, at least two of the six issues mentioned above (i.e., 3.1.1 & 3.1.3), when solved, could help by ensuring a trackable system. Since tracing involves the knowledge, capability, and awareness of the system and its limitations [44], addressing (but not limited to) the remaining four issues could help make a system traceable.…”