Modeling and learning behaviors

摘要: In order to safely navigate in a dynamic environment, robot requires know how the objects populating it will move future. Since this knowledge is seldom available, necessary resort motion prediction algorithms. Due difficulty of modeling various factors that determine (e.g., internal state, perception), often done by applying machine-learning techniques build statistical model, using as input collection trajectories array through sensor camera, laser scanner), and then model predict further motion. This section describes basic concepts involved current learning approaches. After introducing Bayes filter, discusses Growing Hidden Markov Models, an approach which able perform lifelong learning, continuously updating its more data are available. experimental evaluation against two other state-of-the-art approaches, presented consistently outperforms them regarding both accuracy parsimony. The concludes with overview challenges future research directions for