UtilityFlowSim

UtilityFlowSim is a stochastic, utility-based multi-agent traffic simulation for an urban intersection. It simulates mixed traffic participants:

• vehicles
• pedestrians
• bikes
• drones

Agents choose their next movement by maximizing a utility objective that balances route/heading alignment, speed preference, path adherence (lane/sidewalk/altitude behavior), and collision-risk penalties.

What the simulation does

• Builds an intersection environment (including roads, sidewalks, and optional bike-lane setting).
• Spawns initial agents by type based on configuration.
• Optionally generates additional agents during runtime at fixed intervals.
• Advances the world in discrete time steps (dt) until total_simulation_time.
• Runs synchronous decision-making each step so all agents react to the same snapshot.
• Stops updating agents once they reach their destination.
• Produces visualizations, optional animation, and optional CSV trajectory export.

Inputs

The main inputs are defined in config.py inside the CONFIG dictionary.

1) Scenario and environment

• environment_setting: selects layout variant (1 or 2).
• bike_behavior_mode: bike routing mode ("road" or "sidewalk"; in setting 2 this maps to bike-lane behavior).
• Geometry inputs such as lane_width, sidewalk_width, bike_lane_width, and lane counts.
• Drone infrastructure inputs such as num_drone_stations and drone_station_placement_offset.

2) Agent demand (how many agents enter)

• initial_agent_counts: number of each type spawned at time 0.
• enable_dynamic_generation: enables ongoing spawning during simulation.
• dynamic_generation_counts: number of each type spawned per generation event.
• agent_generation_interval: seconds between spawn events.

3) Timing and reproducibility

• dt: simulation step size in seconds.
• total_simulation_time: total duration in seconds.
• random_seed: fixed seed for reproducible runs.

4) Agent dynamics and safety

• Type-specific size/speed/destination-threshold inputs:
  • vehicle_*, bike_*, pedestrian_*, drone_*
• Motion limits:
  • max_acceleration, max_deceleration, max_steering_angle_rad
• Interaction distances:
  • safety-distance parameters like vv_safety_distance, vp_safety_distance, dd_safety_distance, etc.

5) Utility-model parameters

Utility scoring is controlled by weights and shaping parameters such as:

• w_c, w_ell, w_ell_drone
• S_theta, eta_i, S_v, xi_i, S_d, gamma
• beta, epsilon
• collision uncertainty terms (collision_sigma_x/y/z)
• local speed-field terms (speed_influence_*)

These parameters control tradeoffs between efficiency, comfort/path-following, and collision avoidance.

6) Output controls

• show_live_simulation: live plot during simulation.
• create_animation: generate animated output after simulation.
• plot_trajectories, plot_trajectories_with_ids, plot_individual_trajectories
• enable_trajectory_smoothing
• export_to_csv
• animation_output_path: output folder for generated files.

How to run

From the project root:

python main.py

Before running, edit config.py to choose:

• environment mode
• number/type of agents
• total duration and step size
• whether to export plots/animation/CSV

Outputs

Depending on configuration, files are written to animation_output_path (default: ./intersection_plots/), including:

• trajectory plots for ground agents and drones
• optional ID-labeled and individual-agent plots
• optional smoothed trajectory plots
• optional animation
• optional CSV export of trajectories (full_simulation_trajectories.csv)

Notes

• If show_live_simulation is False, the project runs in save/analysis mode and generates configured artifacts after the simulation loop.
• Dynamic generation can substantially increase total simulated agent count during long runs.