Behavior Prediction Based on Obstacle Motion Patterns in Dynamically Changing Environments

This paper proposes a behavior prediction method for navigation application in dynamically changing environments, which predicts obstacle behaviors based on learned obstacle motion patterns (OMP) from observed obstacle motion trajectories. A multi-level prediction model is then proposed that predicts long-term or short-term obstacle behaviors. Simulation results show that it works well in a complex environment and the prediction is consistent with actual behaviors.