Jump to content

Connect SuperML | Leeroopedia MCP: Equip your AI agents with best practices, code verification, and debugging knowledge. Powered by Leeroo — building Organizational Superintelligence. Contact us at founders@leeroo.com.

Implementation:Isaac sim IsaacGymEnvs Runner Play Mode

From Leeroopedia
Sources Domains Last Updated
IsaacGymEnvs, train.py Inference, Integration 2026-02-15 00:00 GMT

Overview

The Runner play-mode dispatch and player factory registration that enables inference execution in IsaacGymEnvs.

Description

When test=True, the launch_rlg_hydra function invokes the Runner with play=True. The Runner uses its player_factory to instantiate the appropriate Player class. Custom player types are registered during build_runner() to extend rl_games with IsaacGymEnvs-specific inference behavior.

Usage

Triggered automatically when the CLI includes test=True. The player factory selection is determined by the algorithm name in the training config (e.g., a2c_continuous, amp_continuous, sac).

Code Reference

Source Location: Repository: NVIDIA-Omniverse/IsaacGymEnvs, File: isaacgymenvs/train.py (L210-215)

Play Dispatch:

# In launch_rlg_hydra, when cfg.test is True:
runner.run({
    'train': False,
    'play': True,
    'checkpoint': cfg.checkpoint,
    'sigma': cfg.sigma
})

Player Factory Registration (in build_runner):

def build_runner(algo_observer):
    runner = Runner(algo_observer=algo_observer)

    # Register custom IsaacGymEnvs player types
    runner.player_factory.register_builder(
        'amp_continuous',
        lambda **kwargs: amp_players.AMPPlayerContinuous(**kwargs)
    )

    # Register custom agent types (for training)
    runner.algo_factory.register_builder(
        'amp_continuous',
        lambda **kwargs: amp_continuous.AMPAgent(**kwargs)
    )

    return runner

I/O Contract

Inputs:

Input Type Description
Configured Runner Runner Runner instance with registered player factories from build_runner()
checkpoint path str Path to .pth file containing trained model weights
sigma float or None Optional exploration noise override
algorithm name str From training config; selects which Player factory to use

Outputs:

Output Type Description
Instantiated Player CommonPlayer / AMPPlayerContinuous / etc. Player with loaded model weights, network in eval mode
Environment VecTask Environment instance ready for rollout evaluation

Player Factory Mapping

Algorithm Name Player Class Description
a2c_continuous CommonPlayer Standard continuous-action PPO player
amp_continuous AMPPlayerContinuous AMP-specialized player with discriminator support
sac (rl_games default) Soft Actor-Critic player from rl_games

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment