Talk - Adaptive Trajectory Prediction Based on Gaussian Stochastic Processes
Abstract
Trajectory prediction is an essential part of many algorithms in autonomous driving. Among its most important applications are behavior planning, collision mitigation, lane keeping and many others. By the stochastic modeling of dynamical parameters of motion the prediction of future trajectories becomes possible. In this document, a regression model using a stochastic approach is developed to successfully predict traffic participant trajectories, and equally important, estimate the prediction’s uncertainty. As the statistical properties of trajectory prediction depend on the traffic scene under consideration, the model is capable of adapting itself to the environment. The model was analyzed on a 400 km long measurement to verify its assumptions.
