A Blockchain Based Liability Attribution Framework for Autonomous Vehicles

The advent of autonomous vehicles is envisaged to disrupt the auto insurance liability model. Unlike the current model, wherein in the event of an accident, the liability is largely attributed to the driver, autonomous vehicles necessitate the consideration of other entities in the automotive ecosystem such as the auto manufacturer, software provider, service technician and the vehicle owner for liability attribution. The proliferation of sensors and connecting technologies in autonomous vehicles enables an autonomous vehicle to gather sufficient data for liability attribution, yet increased connectivity exposes the vehicle to attacks from interacting entities such as an automaker with significant understanding of the inner workings of the autonomous vehicle or the vehicle owner with unrestricted access to the internal network of the vehicle. These possibilities motivate potential liable entities to repudiate their involvement in a collision event to evade liability. While the data collected from vehicular sensors and vehicular communications is an integral part of the evidence for arbitrating liability in the event of an accident, there is also a need to record all interactions between the aforementioned entities to identify potential instances of negligence that may have played a role in the accident. Furthermore, autonomous vehicle sensors are envisaged to collect significant amount of data including personal identifiable information about a vehicle owner which threatens his privacy. In this paper, we propose a BlockChain (BC) based framework that integrates the concerned entities in the liability model and provides untampered evidence for liability attribution and adjuducation. We first describe the liability attribution model, identify key requirements for our proposed BC-based framework and describe the adversarial capabilities of entities. Also, we present a detailed description of relevant data contributing to evidence and further describe how they are stored and used as evidence. Our framework uses permissioned BC, to restrict access to relevant entities and partitions the BC to further tailor data access to relevant BC participants. Finally, we conduct a security analysis to verify that the identified requirements are met and resilience of our proposed framework to identified attacks. Introduction and Motivation The increased adoption of autonomous vehicles is expected to disrupt the auto insurance industry. Vehicle autonomy shifts some or all driving decisions from the human driver to the vehicle, which will require a complete overhaul of existing liability models. Autonomous vehicles are instrumented with a wide array of sensors, embedded computers and communicating technologies to enable better perception of the environment and facilitate independent decision making to avert road transportation hazards. However, the possibility to make independent driving decisions introduces the challenge of liability attribution. A realistic scenario is described in [1] where an autonomous vehicle driver was killed in an accident. The appropriation of liability in this scenario was a significant challenge because the auto manufacturer claimed it did not receive any log about the incident and could not ascertain the drive mode of the vehicle at the time of the accident while the family of the deceased claimed otherwise. Whereas current liability attribution usually 1 ar X iv :1 80 2. 05 05 0v 1 [ cs .C R ] 1 4 Fe b 20 18 assigns blame to the driver, liability for autonomous vehicles is to be shared between multiple entities responsible for the operation of the vehicle including the auto manufacturer, maintenance service provider, driver and software provider. The possibility of shared liability therefore highlights the urgent need for a new auto insurance liability framework. The authors in [2] introduce the following two categories of liability attribution for autonomous vehicles: i) Product liability which refers to damages due to product defects such as design failure and manufacturing failures. Liability is either attributed to the auto manufacturer when an accident occurs and the vehicle is in autonomous mode or to the software provider when a software program is deemed to have led to the accident by causing erroneous action. A service technician is liable if an accident is traced to its last action on the vehicle. ii) Negligence liability which refers to damages due to neglects to execute an action. Liability is attributed to a vehicle owner when he fails to execute an instruction (software update) from an auto manufacturer or software provider. An auto manufacturer is however considered negligent if the accident could have been prevented by a human driver. Therefore, given the associated liability costs, e.g., compensation and potential tarnishing of reputation, an entity that is deemed liable may be strongly motivated to deny its actions thus making liability attribution challenging. Also, an autonomous vehicle houses multiple data gathering sensors such as Light Detection and Ranging (LIDAR) for generating a 3D map of its surroundings, cameras to read speed limit signs and watch lane marks to prevent drifting, GPS and tire pressure monitoring sensors. Given the high exposure of the autonomous vehicle to external communication, an auto manufacturer with significant understanding of the inner workings of the vehicle could remotely tamper with sensors generating data for evidence to evade liability. Remote exploitation has already been successfully demonstrated for a modern day vehicle [3]. An autonomous vehicle is envisaged to generate significant amount of data, some of which could reveal private information about the driver and passenger [2]. This introduces privacy concerns when data generated for evidence is either retrieved or transmitted to decision makers for dispute settlement. In addition, gathering data that constitutes evidence for liability attribution is challenging. The National Road and Motorists Association (NRMA) [4], Australia identified data constituting evidence as time of event, date, location, description of event, witness testimonies and pictorial evidence. In the traditional liability model, when an accident occurs, forensic investigators visit the site to obtain physical evidence such as impact of damage, vehicle position and vehicle heading. They also extract data from the Event Data Recorder (EDR); a black-box installed in the vehicle that stores pre and post collision data such as speed, accelerator angle, usage of safety systems etc. These data enable a vehicle to prove its recent behaviour for liability purposes. However, such data is not sufficient for liability attribution as it cannot be used to prove the interaction between a vehicle and other potentially liable entities to reveal potential instances of negligence which could also be the cause of the accident. Thus, new mechanisms are needed to record relevant data exchanges between liable entities as evidence to ascertain if the accident was due to negligence. Furthermore, given the popular consensus to attribute liability to an auto manufacturer or software provider for product defects and to the vehicle owner for negligence [2], a likely liable entity might be motivated to deny the availability of data or receipt of data contributing to evidence. In the realistic scenario described in [1], the auto manufacturer could not ascertain the cause of an accident because the accident log files were not logged in their server due to severity of damage. The auto manufacturer was also denied access to more information by the vehicle owner to aid its investigation. The consensus to split liability [2] in the autonomous vehicle setting and the vested interest of liable entities in the outcome of liability decisions, highlight the need for a transparent liability framework where such data contributing to evidence is accessible and universally accepted by concerned entities in the liability framework for expediting liability decisions. The emerging Blockchain technology has the potential to underpin a new liability framework for autonomous vehicles as it provides trustless consensus. Blockchain (BC) is an immutable and distributed ledger technology that provides verifiable record of transactions in the form of an interconnected series of data blocks. Blockchain’s immutability is achieved by using the hash of the previous block to chain interconnected data blocks so that changing the content of a single block will require changing the headers of previous blocks up to the genesis block; the first block in the chain. In addition to the immutability and security features of BC, it also offers privacy to nodes in a network and therefore is a useful technology to address aforementioned