Multiple environments#

IR-SIM supports creating and running multiple environments simultaneously. Each environment maintains its own isolated state, including world parameters, objects, and simulation time. This is useful for:

Parallel simulations: Running multiple scenarios simultaneously
Comparison studies: Comparing different algorithms or configurations
Training: Running multiple instances for reinforcement learning

Creating Multiple Environments#

import irsim

# Create two separate environments
env1 = irsim.make('scenario_a.yaml')
env2 = irsim.make('scenario_b.yaml')

# Each environment has its own state
print(f"Env1 robots: {env1.robot_number}")
print(f"Env2 robots: {env2.robot_number}")

Isolated Parameters#

Each environment has completely separate parameter instances:

import irsim

env1 = irsim.make('world1.yaml')
env2 = irsim.make('world2.yaml')

# World parameters are isolated
env1.world_param.control_mode = 'keyboard'
print(f"Env1 control mode: {env1.world_param.control_mode}")  # keyboard
print(f"Env2 control mode: {env2.world_param.control_mode}")  # auto (unchanged)

# Simulation time is independent
for _ in range(10):
    env1.step()

for _ in range(5):
    env2.step()

print(f"Env1 time: {env1.time}")  # 1.0 (10 steps * 0.1)
print(f"Env2 time: {env2.time}")  # 0.5 (5 steps * 0.1)

Running Multiple Environments#

Sequential Execution

import irsim

env1 = irsim.make('scenario_a.yaml', display=True)
env2 = irsim.make('scenario_b.yaml', display=True)

# Run both environments in the same loop
for i in range(500):
    # Step both environments
    env1.step()
    env2.step()

    # Render both (creates two windows)
    env1.render(0.01)
    env2.render(0.01)

    # Check completion independently
    if env1.done() and env2.done():
        break

env1.end()
env2.end()

Headless Parallel (for Training)

import irsim

# Create multiple headless environments for training
num_envs = 4
envs = [
    irsim.make(f'training_world.yaml', display=False, seed=i)
    for i in range(num_envs)
]

# Run training loop
for episode in range(100):
    # Reset all environments
    for env in envs:
        env.reset()

    # Collect experiences from all environments
    for step in range(1000):
        actions = [get_action(env) for env in envs]  # Your policy

        for env, action in zip(envs, actions):
            env.step(action)

        # Check if any environment is done
        if all(env.done() for env in envs):
            break

# Clean up
for env in envs:
    env.end()

Parameter Isolation Details#

Each environment instance creates and binds its own parameter objects:

world_param - Simulation state and settings:

Attribute	Type	Default	Description
`time`	`float`	`0.0`	Current simulation time
`control_mode`	`str`	`"auto"`	Control mode (`"auto"` or `"keyboard"`)
`collision_mode`	`str`	`"stop"`	Collision handling mode
`step_time`	`float`	`0.1`	Time step duration (seconds)
`count`	`int`	`0`	Simulation step counter

env_param - Environment objects and utilities:

Attribute	Type	Description
`objects`	`list[ObjectBase]`	List of all objects in the environment
`logger`	`EnvLogger`	Logger instance for this environment
`GeometryTree`	`STRtree`	Spatial index for collision detection
`platform_name`	`str`	Operating system name

path_param - File path management:

Attribute	Type	Description
`root_path`	`str`	Path to irsim package directory
`ani_buffer_path`	`str`	Path for animation frame buffer
`ani_path`	`str`	Path for saved animations
`fig_path`	`str`	Path for saved figures

Objects within each environment reference their own environment’s parameters:

Note

When creating multiple environments with display=True, each environment opens its own visualization window. For training or batch simulations, use display=False to disable rendering and improve performance.