Human driving models aim at producing human-like driving strategies by mimicking the behavior of drivers. Drivers optimize several objectives when traveling along a route, such as the traveling time and the fuel consumption. However, these objectives are not taken into account when building human driving models. To overcome this shortcoming, we designed a two-level Multiobjective Optimization algorithm for discovering Human-like Driving Strategies (MOHDS) that combines the human driving models with the optimization of the traveling time and the fuel consumption. Consequently, MOHDS enables to simultaneously mimic human driving behavior and optimize relevant driving objectives. MOHDS was tested on a two-lane rural route and compared to the existing approaches for human driving modeling. The results show that, unlike the existing approaches, MOHDS finds the driving strategies with various tradeoffs between the objectives.
COBISS.SI-ID: 30649383
This article describes a new architecture for archiving the data that are highly structured. This architecture allows a complete parametrization of the data model, including the set of database tables, relations among them, etc. In addition, it also enables the replication of data, i.e., the data from the database, as well as additional data stored in (larger) files and connected with the database. This architecture has been initially developed for astronomical data, and afterwards significantly upgraded and enhanced during this project for the purpose of collecting data on driving from both the algorithms and human drivers.
COBISS.SI-ID: 31187239
This paper presents a methodology for evaluation of driving performance based on speeding, acceleration, lane control and safety distance. All these variables are measured in a motion-based driving simulator. We report on a user study in which we obtained the proposed variables for 29 drivers. The results enable us to propose a general evaluation score of driving performance, which can be used for profiling driver behavior.
COBISS.SI-ID: 11742548
When drivers drive along the route, they optimize several objectives, e.g., the traveling time and the fuel consumption. However, these objectives are usually not taken into account when designing human driving models. For the purpose of the simultaneous optimization of both the human aspects and the driving objectives, we developed a Multiobjective optimization algorithm for discovering human-like driving strategies (MOHDS). This algorithm includes human driving models and optimizes three objectives: the traveling time, the fuel consumption and the similarity with human driving. MOHDS was evaluated on three routes that included bends, slopes, other vehicles, and a highway. The obtained driving strategies were compared with human driving strategies. The results indicate that MOHDS finds driving strategies that are comparable to human driving strategies when taking into account the driving objectives in most of the test scenarios.
COBISS.SI-ID: 30854695
This paper introduces a methodology for acquisition and evaluation of driving data and analysis of overall performance. We report on a user study in which we obtained driving performance data in a motion based driving simulator. The obtained data is based on safety distance, speeding, winding and aggressive acceleration. The results of the user study enable us to perform profiling of drivers based on their driving behavior and with the aid of a scoring system.
COBISS.SI-ID: 11822932