Real time level lane decision algorithm based on autonomous vehicle sensor data

Real Time autonomous vehicle localization is considered. A conceptual framework defining the vehicle lane positioning on an arbitrary road network is introduced. The proposed model-based feedback scheme, receiving as main inputs a recent observation set and the current vehicle state, returns the vehicle level lane position. Various decision-making criteria are conceived each is associated with low algorithmic complexity and thus reduced computational cost. A prioritization criteria policy is defined selecting the most appropriate constraints to be satisfied at each decision time. Since a minimum number of trustworthy, sensor available information is exploited the need for additional, costly devices is substantially decreased. Depending upon the reference vehicle context and data reliability no solution or even multiple responses may be implied in which case the more reliable answer should be selected. Alternatives to compensate the case of no lane identification are proposed. The algorithm performance is evaluated and its effectiveness is illustrated in both qualitative and quantitative terms. Complex ground truth are utilized incorporating multiple traffic conditions and types of road infrastructure. Metrics on the correctness and precision of the suggested scheme, estimation failure, entry validation etc. are presented.