In this thesis, motion planning for automated vehicles considering uncertainties and interactions with human drivers is addressed for lane merging in dense traffic. Here, two main problems arise: 1. A model for the reaction of human drivers is necessary, and 2. the search space increases drastically which prohibits many planning techniques. This is addressed by training models based on neural networks and using them to focus a MCTS search process to promising regions of the state space. Based on real scenarios, the planner showed an increased lane change success rate compared to a baseline. A nonlinear optimization improved the high-level trajectory in comfort and safety considering the capabilities a controller.
«
In this thesis, motion planning for automated vehicles considering uncertainties and interactions with human drivers is addressed for lane merging in dense traffic. Here, two main problems arise: 1. A model for the reaction of human drivers is necessary, and 2. the search space increases drastically which prohibits many planning techniques. This is addressed by training models based on neural networks and using them to focus a MCTS search process to promising regions of the state space. Based on...
»