Implementation:Facebookresearch Habitat lab PPOTrainer eval checkpoint

Knowledge Sources	Habitat-Lab
Domains	Hierarchical_RL, Evaluation
Last Updated	2026-02-15 02:00 GMT

Overview

Concrete checkpoint evaluation method for both flat and hierarchical policies, delegating to HabitatEvaluator for episode execution and metric collection.

Description

The PPOTrainer._eval_checkpoint method loads a policy checkpoint, creates evaluation environments, initializes the evaluator, and runs the agent through evaluation episodes. For hierarchical policies, it loads the full HierarchicalPolicy with all skill checkpoints. Rearrangement-specific metrics (composite_success, rearrangement_success, force metrics) are collected alongside standard navigation metrics.

Usage

Called with --run-type eval from the command line, or programmatically by passing a checkpoint path.

Code Reference

Source Location

Repository: habitat-lab
File: habitat-baselines/habitat_baselines/rl/ppo/ppo_trainer.py
Lines: L803-902 (_eval_checkpoint), L294-328 (save_checkpoint)

Signature

class PPOTrainer(BaseRLTrainer):
    def _eval_checkpoint(
        self,
        checkpoint_path: str,
        writer: TensorboardWriter,
        checkpoint_index: int = 0,
    ) -> None:
        """
        Evaluate a single checkpoint.

        Args:
            checkpoint_path: Path to .pth checkpoint file
            writer: TensorBoard writer for logging
            checkpoint_index: Index for naming in logs
        """

Import

from habitat_baselines.rl.ppo.ppo_trainer import PPOTrainer

I/O Contract

Inputs

Name	Type	Required	Description
checkpoint_path	str	Yes	Path to trained checkpoint file
writer	TensorboardWriter	Yes	Writer for logging metrics
checkpoint_index	int	No	Index for log naming (default 0)

Outputs

Name	Type	Description
Metrics	Dict[str, float]	Aggregated rearrangement metrics logged to writer
Videos	.mp4 files	Optional evaluation recordings

Usage Examples

Evaluate HRL Checkpoint

python -u habitat-baselines/habitat_baselines/run.py \
    --exp-config habitat-baselines/habitat_baselines/config/rearrange/rl_hierarchical.yaml \
    --run-type eval \
    habitat_baselines.eval.video_option='["disk"]' \
    habitat_baselines.eval_ckpt_path_dir=checkpoints/rl_hierarchical.pth

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment