Simulating diverse long-term future trajectories in road scenes

A non-transitory computer-readable storage medium comprising a computer-readable program for simultaneous multi-agent recurrent trajectory prediction, wherein the computer-readable program when executed on a computer causes the computer to perform the steps of:

• reconstructing, via a scene generator, a topological layout of a scene from a dataset including real-world data;
• generating a road graph of the scene, the road graph capturing a hierarchical structure of interconnected lanes;
• incorporating vehicles from the scene on the generated road graph by utilizing tracklet information available in the dataset;
• assigning the vehicles to their closest lane identifications;
• identifying diverse plausible behaviors for every vehicle in the scene;
• sampling one behavior from the diverse plausible behaviors, by a diverse behavior generator, for every vehicle in the scene;
• sampling a velocity profile using the sampled behavior, by the diverse behavior generator, to identify a closest behavior from real scenes using heuristics with nearest neighbor sampling for every vehicle in the scene;
• obtaining realistic trajectory characteristics for every vehicle in the scene, by the diverse behavior generator, by associating the sampled velocity profile with a plurality of real-world trajectory samples that are classified into different maneuver categories using a set of centerlines from the road graph;
• generating a plurality of simulated diverse trajectories using a dynamics simulator by utilizing the road graph, the sampled velocity profile, and the realistic trajectory characteristics with a desired destination;
• and
• outputting on a visualization device to allow a user to select one or more of the plurality of simulated diverse trajectories for recreating a desired driving scenario.

A system for simultaneous multi-agent recurrent trajectory prediction, the system comprising:

• a memory;
• and
• one or more processors in communication with the memory configured to:
• reconstruct, via a scene generator, a topological layout of a scene from a dataset including real-world data;
• generate a road graph of the scene, the road graph capturing a hierarchical structure of interconnected lanes;
• incorporate vehicles from the scene on the generated road graph by utilizing tracklet information available in the dataset;
• assign the vehicles to their closest lane identifications;
• identify diverse plausible behaviors for every vehicle in the scene;
• sample one behavior from the diverse plausible behaviors, by a diverse behavior generator, for every vehicle in the scene;
• sample a velocity profile using the sampled behavior, by the diverse behavior generator, to identify a closest behavior from real scenes using heuristics with nearest neighbor sampling for every vehicle in the scene;
• obtain realistic trajectory characteristics for every vehicle in the scene, by the diverse behavior generator, by associating the sampled velocity profile with a plurality of real-world trajectory samples that are classified into different maneuver categories using a set of centerlines from the road graph;
• generate a plurality of simulated diverse trajectories using a dynamics simulator by utilizing the road graph, the sampled velocity profile, and the realistic trajectory characteristics with a desired destination;
• and
• output on a visualization device to allow a user to select one or more of the plurality of simulated diverse trajectories for recreating a desired driving scenario.