Simulation

URBAN-SIM supports both synchronous and asynchronous multi-environment simulations. These two modes differ in how environments are spawned.

⏱️ Synchronous Simulation

In synchronous mode, all environments are advanced together in lock-step. They share the same scene layout, including map structure, object placement, and pedestrian routes. This setting is ideal for:

• Controlled debugging

• Frame-aligned evaluation

• Consistent benchmarking across environments

python urbansim/envs/separate_envs/random_env.py --num_envs 16 --scenario_type {clean,static,dynamic}

Example Scenario Types:

• clean: Empty environment without obstacles

• static: With static obstacles

• dynamic: With both static and dynamic (e.g., pedestrians) elements

⚡ Asynchronous Simulation

In asynchronous mode, each environment is generated independently, allowing for diverse scenario layouts, object instances, appearances, etc. This is especially useful for improving robustness, or simulating real-world noise.

python urbansim/envs/separate_envs/random_env.py --num_envs 16 --scenario_type {clean,static,dynamic} --use_async

Advantages:

• Enables greater variation in observation-action loops

• Closer to real-world multi-agent or multi-robot setups

• Can increase training speed under high environment counts

🔄 Semantic Differences: Synchronous vs. Asynchronous Scenarios

In synchronous simulation, all environments share the same scene layout, obstacle placement, and map structure. This is ideal for evaluation and debugging where controlled consistency is needed.

In asynchronous simulation, each environment instance is procedurally varied. This allows:

• Diverse layouts and building placements

• Varying pedestrian paths and densities

• Scene-level randomness in lighting or asset combinations

This diversity helps improve the generalization and robustness of navigation policies.

Note

✅ Asynchronous simulation is ideal for large-scale training across diverse environments.

🖼️ Scenario Visualizations

Different scenario types can be combined with both simulation modes. For example:

• clean (empty world)

• static (with immovable structures)

• dynamic (with moving pedestrians)

Sync - Clean	Sync - Static	Sync - Dynamic

Async - Clean	Async - Static	Async - Dynamic