Development of a Real-time strategical and tactical decision making model for non-cooperative autonomous vehicles

Kurzfassung

In this thesis a strategical and tactical decision making model working real- time in a noncooperative scenario for an autonomously driving vehicle is presented. Firstly, to model a lane change behavior it is necessary to model the drivers’ acceleration behavior and this is determined by modelling a car- following model. A Foresighted Driver Model (FDM) is implemented to model the driver behavior in a car- following regime. As for the two components of an FDM, risk and utility, risk is modelled using is quadratic penalty function based on the indicator Time- To- Closest- Encounter (TTCE) is developed. Utility term is the free driving acceleration term is based on the current velocity and desired velocity based on the vehicle in front. Risk function is then generated for different possible velocities within the range defined. A risk- map is generated to give a perspective of how risk varies with the velocity. A non- linear solver based on sequential quadratic programming (SQP) is used to keep the risk minimal. Risk function is modelled for two kinds of risks, Risk due to Collision and Risk due to road curvature. The lane changing behavior modelled is based on a target gap selection model. Gap selection is done based on a utility function which takes into account the gap characteristics (ex: length, distance to the gap and its effectiveness) is modelled as a logit model. The gap with highest utility is selected as the best gap to merge in. Finally, this is implemented together with FDM to model the driver behavior in case of car- following regime and lane changing behavior and tested in several scenarios (presented in latter sections) in simulation and also on a real vehicle at a closed test field only for the case of car- following model. Algorithms were implemented and simulated is C++, vehicle used for tests is “ViewCar2” of the Institute of Transportation Systems at DLR, Braunschweig