[GitHub Code for this Chapter: chapter2_3 branch](https://github.com/Lab-of-AI-and-Robotics/IsaacLab-Tutorial/tree/chapter2_3)

Objective

To deconstruct the ManagerBasedRLEnvCfg class, the declarative "blueprint" that defines our entire reinforcement learning task. This chapter delves into the most critical file in our project: the environment configuration (_env_cfg.py). By examining the default test project's example, we will gain a deep understanding of how to define a scene, and how the managers for actions, observations, rewards, and events are configured.

3.1. The Blueprint for Behavior: _env_cfg.py

In the last chapter, we established that the Manager-Based workflow separates configuration (what to do) from implementation (how to do it). We now shift our focus to that configuration.

Navigate to your project's task folder: .../test/source/test/test/tasks/manager_based/test/. Inside, you'll find test_env_cfg.py. Think of this file as a detailed recipe. It tells Isaac Lab's "engine" (ManagerBasedRLEnv) exactly what ingredients to use (a robot, a ground plane) and the rules for success (the rewards and terminations). Our job as researchers is primarily to be the "chef" who fine-tunes this recipe.

Let's dissect the TestEnvCfg class to understand its main components.

3.2. Dissecting the TestEnvCfg Class

Opening test_env_cfg.py reveals the main TestEnvCfg class. This top-level class acts as a container, assembling all the individual configuration modules (SceneCfg, ActionsCfg, etc.) into a single, cohesive definition for our environment. This modular assembly is the core principle of the Manager-Based workflow.

The code below shows how these components are brought together. We will then break down each of these modules one by one.

@configclass
class TestEnvCfg(ManagerBasedRLEnvCfg):
    # Scene settings
    scene: TestSceneCfg = TestSceneCfg(num_envs=4096, env_spacing=4.0)
    # Basic settings
    observations: ObservationsCfg = ObservationsCfg()
    actions: ActionsCfg = ActionsCfg()
    events: EventCfg = EventCfg()
    # MDP settings
    rewards: RewardsCfg = RewardsCfg()
    terminations: TerminationsCfg = TerminationsCfg()

    # Post initialization
    def __post_init__(self) -> None:
        """Post initialization."""
        # general settings
        self.decimation = 2
        self.episode_length_s = 5
        # viewer settings
        self.viewer.eye = (8.0, 0.0, 5.0)
        # simulation settings
        self.sim.dt = 1 / 120
        self.sim.render_interval = self.decimation

Notice the __post_init__ method. This special function runs after all the configuration objects are initialized, allowing us to set global parameters like episode_length_s and the simulation timestep sim.dt.

SceneCfg (The Stage)

This object defines the physical world. As seen in the TestSceneCfg class, it defines a ground plane, spawns the robot using a pre-defined CARTPOLE_CFG, and sets up lighting. It also sets crucial parameters like num_envs and env_spacing in the main TestEnvCfg.

@configclass
class TestSceneCfg(InteractiveSceneCfg):
    """Configuration for a cart-pole scene."""

    # ground plane
    ground = AssetBaseCfg(
        prim_path="/World/ground",
        spawn=sim_utils.GroundPlaneCfg(size=(100.0, 100.0)),
    )

    # robot
    robot: ArticulationCfg = CARTPOLE_CFG.replace(prim_path="{ENV_REGEX_NS}/Robot")

    # lights
    dome_light = AssetBaseCfg(
        prim_path="/World/DomeLight",
        spawn=sim_utils.DomeLightCfg(color=(0.9, 0.9, 0.9), intensity=500.0),
    )

ActionsCfg (What the Agent Can Do)

This defines the agent's action space. In this case, joint_effort allows the policy to apply a force (effort) to the slider_to_cart joint, scaled by 100.0.

@configclass
class ActionsCfg:
    """Action specifications for the MDP."""

    joint_effort = mdp.JointEffortActionCfg(asset_name="robot", joint_names=["slider_to_cart"], scale=100.0)