PREDICTIVE POTENTIAL FIELD-BASED COLLISION AVOIDANCE FOR MULTICOPTERS
Keywords: UAVs, Obstacle Avoidance, Motion Model Learning, Potential Field Method
Abstract. Reliable obstacle avoidance is a key to navigating with UAVs in the close vicinity of static and dynamic obstacles. Wheel-based mobile robots are often equipped with 2D or 3D laser range finders that cover the 2D workspace sufficiently accurate and at a high rate. Micro UAV platforms operate in a 3D environment, but the restricted payload prohibits the use of fast state-of-the-art 3D sensors. Thus, perception of small obstacles is often only possible in the vicinity of the UAV and a fast collision avoidance system is necessary. We propose a reactive collision avoidance system based on artificial potential fields, that takes the special dynamics of UAVs into account by predicting the influence of obstacles on the estimated trajectory in the near future using a learned motion model. Experimental evaluation shows that the prediction leads to smoother trajectories and allows to navigate collision-free through passageways.