Driver Intention-based Vehicle Threat Assessment using Random Forests and Particle Filtering

Abstract

One of the key technologies to safely operate self-driving vehicles is the threat assessment of other vehicles in the neighborhood of a self-driving vehicle. Threat assessment algorithms must be capable of predicting the future movement of other vehicles. Many algorithms, however, predict future trajectories based only on the model of the dynamics and the environment, which implies that they sometimes make too conservative predictions. This work reduces this conservativeness by capturing the driver intention of other vehicles using a random-forests classifier. Then, the algorithm computes possible future trajectories with a sequential Monte Carlo method, which biases the predicted trajectory by the recognized intention. Lastly, the algorithm calculates the potential threat to the ego vehicle. To evaluate the performance, we conduct numerical simulations and show that the proposed algorithm can accurately capture driver intentions and prevent motion predictions that are too conservative.