Spatiotemporal Learning of Multivehicle Interaction Patterns in Lane-Change Scenarios

Interpretation of common-yet-challenging interaction scenarios can benefit well-founded decisions for autonomous vehicles. Previous research achieved this using their prior knowledge specific with predefined models, limiting adaptive capabilities. This paper describes a Bayesian nonparametric approach that leverages continuous (i.e., Gaussian processes) and discrete Dirichlet stochastic processes to reveal underlying patterns the ego vehicle other nearby Our model relaxes dependency on number surrounding vehicles by developing an acceleration-sensitive velocity field based processes. The experiment results demonstrate represent spatial interactions between its surroundings. Then, model, integrating hidden Markov is developed learn over temporal space segmenting clustering sequential data into interpretable granular automatically. We then evaluate our in highway lane-change highD dataset collected from real-world settings. Results proposed provides insight complicated multiple traffic participants transition properties relationships. sheds light efficiently analyzing kinds multi-agent interactions, such as vehicle-pedestrian interactions. View demos via https://youtu.be/z_vf9UHtdAM.