Humanoids and Intelligence Systems Lab - Institute for Anthropomatics and Robotics

Behavior Planning

behavior planning

The subject of "Behavior Planning" is the research of planning methods, which can be used to implement an anticipatory and self-aware decision making. The usual approach for decision making for autonomous cars is to manually model decisions for different environment states. In contrast to this, we want to employ POMDPs to make the car conclude the optimal decision for every probability distribution over the state space automatically. This method enables the vehicle to explicitly consider its own uncertainty and lack of knowledge, resulting from incomplete perception. One of the main points of this research is the integration of empirical knowledge the car gains in real or simulated driving experiments. With help of a human "driving instructor" the car gathers knowledge about a target-aimed and sufficient discretization of the state space