Unity-Technologies/ml-agents · error · TrainerConfigError

The trainer config contains an unknown trainer type {trainer

Error message

The trainer config contains an unknown trainer type {trainer_settings.trainer_type} for brain {brain_name}

What it means

TrainerFactory._initialize_trainer converts trainer settings and, on KeyError (the trainer_type key not present in its trainer-type-to-class mapping), raises TrainerConfigError stating the trainer type is unknown for the given brain. It is a config-validation error guarding the trainer_type hyperparameter.

Source

Thrown at ml-agents/mlagents/trainers/trainer/trainer_factory.py:116

        min_lesson_length = param_manager.get_minimum_reward_buffer_size(brain_name)

        trainer: Trainer = None  # type: ignore  # will be set to one of these, or raise

        try:
            trainer_type = all_trainer_types[trainer_settings.trainer_type]
            trainer = trainer_type(
                brain_name,
                min_lesson_length,
                trainer_settings,
                train_model,
                load_model,
                seed,
                trainer_artifact_path,
            )

        except KeyError:
            raise TrainerConfigError(
                f"The trainer config contains an unknown trainer type "
                f"{trainer_settings.trainer_type} for brain {brain_name}"
            )

        if trainer_settings.self_play is not None:
            trainer = GhostTrainer(
                trainer,
                brain_name,
                ghost_controller,
                min_lesson_length,
                trainer_settings,
                train_model,
                trainer_artifact_path,
            )
        return trainer

View on GitHub (pinned to 3ecb446f75)

Solutions

1. Set trainer_type to one of the supported values: ppo, sac, or online_bc (or use self_play settings for ghost)
2. Check spelling and casing in the YAML (must be lowercase)
3. Verify trainer_type against the schema for your installed mlagents version (mlagents-learn --help shows valid types)
4. Update ML-Agents if your config targets a newer version with different trainer names

Example fix

// before (config)
trainer_type: PPO2
// after
trainer_type: ppo

VALID = {"ppo", "sac", "online_bc"}
assert trainer_settings.trainer_type.lower() in VALID, f"Unknown trainer_type {trainer_settings.trainer_type}"

def is_known_trainer_type(settings) -> bool:
    return str(settings.trainer_type).lower() in {"ppo", "sac", "ghost", "online_bc"}

from mlagents.trainers.exception import TrainerConfigError
try:
    trainer = trainer_factory.initialize_trainer_and_rl_glue(...)
except TrainerConfigError as e:
    logger.error(f"Fix your YAML: {e}")
    raise SystemExit(1)

When it happens

Trigger: Setting trainer_type in the trainer config to anything other than ppo, sac, ghost (self-play) or online_bc; the settings dict lookup raises KeyError and is re-raised as TrainerConfigError with the brain name.

Common situations: Typo in YAML like trainer_type: PPO (uppercase) or 'ppo2'; using a trainer type removed in the installed ML-Agents version (e.g. il/behavioral cloning renamed to online_bc); copying community configs for a different ML-Agents release.

Related errors

• The schedule {self.schedule} is invalid.
• The trainer was unable to process any of the provided inputs
• The one of the goals uses variable length observations. This
• Trainer was unable to process any of the goals provided as i
• Visual observation resolution ({width}x{height}) is too smal

AI-assisted analysis of Unity-Technologies/ml-agents@3ecb446f75 (2026-09-02). Data as JSON: /api/errors/584376c3fbacf6ea. Report an issue: GitHub.