Simulated Pedestrian Model

The Simulated Pedestrian (SP) model provides testers with realistic intelligent pedestrians to include in test scenarios. The SP model is a hierarchical behaviour model consisting of high-level decision layers that inform low-level layers to produce realistic decisions and human motion in a traffic environment. SP pedestrians dynamically interact with their environment and other road users with data-derived and highly configurable behaviours.

The SP model is an extension of the base GeoScenario language and focuses scenarios around its dynamic pedestrian agents. Scenarios can include both vehicle (optionally using the SDV model) and pedestrian agents. The model requires specific scenario and environmental constraints including a specified GeoScenario representation of pedestrians and pedestrian-specific components of the Lanelet2 map.

The SP model aims to express realistic high-level decision processes and low-level movements as its pedestrians navigate through the traffic environment. High-level decisions are produced by behaviour trees in the Behaviour layer. These decisions are passed to the lower layers of the model responsible for translating them into motion using an adapted Social Force model.

Layered Model Overview

The model uses a layered architecture with a hierarchical structure to plan and execute decisions. Each layer in the model has access to the Traffic State Estimation, a snapshot representation of the current state of the environment.

  • Behaviour Layer: This layer contains the pedestrian's behaviour tree. At each scenario simulation cycle, it traverses the tree and selects a maneuver to be executed by the lower layers. The selected maneuver is then passed to the Maneuver layer for interpretation.
  • Maneuver Layer: This layer receives a maneuver from the Behaviour layer and determines how the pedestrian must adjust their current trajectory to execute the maneuver based on the current state of the environment and the other dynamic agents.
  • Motion Planner Layer: This layer applies an adapted Social Force model (SFM) with the instructions from the Maneuver layer. The SFM outputs the pedestrian's next state information (position, velocity, acceleration, heading) and updates the simulation environment.

Notes:

  • Pedestrian behaviours can be varied with different configurations of behaviour trees. We define three levels of aggressiveness to model different observed behaviours in traffic environments.
  • The model abstracts the pose of the pedestrians and applies a point-mass model representation.

The next sections discuss the details of the model's structure and implementation.