Crowd Navigation : Autonomous navigation in an urban environment

Abstract: In this thesis, strategies for navigating a crowded area using an autonomous holonomic robot are discussed and evaluated. The focus is set on path planning and the topic is therefore largely decoupled from the prediction (i.e. machine learning) and control theory techniques needed for a practical implementation outside of the simulated environment. Existing methods and algorithms for path planning in highly dynamic environments are compared using several measures via computer simulations in different environments. A new, effective, and yet simple, algorithm is introduced and proven to be useful in certain scenarios. This algorithm, ART, predicts the future states of the crowd and using these predictions finds better paths to the goal than traditional algorithms.