Non-Line of Sight Test Scenario Generation for Connected Autonomous Vehicle




Journal Title

Journal ISSN



Volume Title


De Montfort University


Thesis or dissertation

Peer reviewed


Connected autonomous vehicles (CAV) level 4-5 use sensors to perceive their environment. These sensors are able to detect only up to a certain range and this range can be further constrained by the presence of obstacles in its path or as a result of the geometry of the road, for example, at a junction. This is termed as a non-line of sight (NLOS) scenario where the ego vehicle (system under test) is unable to detect an oncoming dynamic object due to obstacles or the geometry of the road. A large body of work now exist which proposes methods for extending the perception horizon of CAV’s using vehicular communication and incorporating this into CAV algorithms ranging from obstacle detection to path planning and beyond. Such proposed new algorithms and entire systems needs testing and validating, which can be conducted through primarily two ways, on road testing and simulation. On-road testing can be extremely expensive and time-consuming and may not cover all possible test scenarios. Testing through simulation is inexpensive and has a better scenario space coverage. However, there is currently a dearth in simulated testing techniques that provides the environment to test technologies and algorithms developed for NLOS scenarios. This thesis puts forward a novel end-to-end framework for testing the abilities of a CAV through simulated generation of NLOS scenarios. This has been achieved through following the development process of Functional, Logical and Concrete scenarios along the V-model-based development process in ISO 26262. The process begins with the representation of the NLOS environment (including the digital environment) knowledge as a scalable ontology where Functional and Logical scenarios stand for different abstraction levels. The proposed new ontology comprises of six layers: ‘Environment’, ‘Road User’, ‘Object Type’, ‘Communication Network’, ‘Scene’ and ‘Scenario’. The ontology is modelled and validated in protégé software and exported to OWL API where the logical scenarios are generated and validated. An innumerable number of “concrete” scenarios are generated as a result of the possible combinations of the values from the domains of each concept’s attributes. This research puts forward a novel genetic- algorithm (GA) approach to search through the scenario space and filter out safety critical test scenarios. A critical NLOS scenario is one where a collision is highly likely because the ego vehicle was unable to detect an obstacle in time due to obstructions present in the line-of-sight of the sensors or created due to the road geometry. The metric proposed to identify critical scenarios which also acts as the GA’s fitness function uses the time-to-collision (TTC) and total stopping time (TST) metric. These generated critical scenarios and proposed fitness function have been validated through MATLAB simulation. Furthermore, this research incorporates the relevant knowledge of vehicle-to-vehicle (V2V) communication technologies in the proposed ontology and uses the communication layer instances in the MATLAB simulation to support the testing of the increasing number of approaches that uses communications for alerting oncoming vehicles about imminent danger, or in other word, mitigating an otherwise critical scenario.





Research Institute