Tracking multiple moving objects in a dynamic environment for autonomous navigation

To deploy autonomous vehicles in dynamic environments its imperative that the robots have a model of the surroundings, with emphasis in the dynamic aspects of the refered surroundings. With the methods presented in this paper the robot is capable of detect and track several moving objects in its surroundings, using only a range finder sensor. The major frameworks that enable the presented method are particle filters and Sample-based Joint Probabilistica Data Association Filters. Particle filters because they are able to predict, with some accuracy, the next state of a non-linear, non-gaussian, multimodal model. And SJPDAFs because they can easely make the data association between sensor data and particle filter that tracks the moving object state (position). Several methods based on occupancy grids are presented to assist the data association. Several experimental results are presented using real data taken from the laser range finder and the robot odometry, demonstrating the effectiveness of the presented methods.