Replay Module¶

The replay module provides replay and visualization of evaluation results, with trajectory rendering, observation rendering, and info panel display.

Overview¶

The replay module is used for:

Trajectory Visualization - Display navigation trajectory on the map
Observation Replay - Replay observation images per step
Info Panel - Show evaluation metrics and status
Video Generation - Generate replay video files

Core Components¶

DataRecorder - Records episode data during evaluation (trajectory, observations, actions)
ReplayLoader - Loads recorded replay data
VideoReplayer - Generates visualization videos combining trajectory and observations
Renderers - trajectory_renderer, observation_renderer, info_panel_renderer

Replay Data Format¶

Replay data is stored in the evaluation results directory:

eval_results/
├── episode_001/
│   ├── replay_data.json      # Replay data
│   ├── trajectory.json        # Trajectory data
│   └── observations/          # Observation images
│       ├── step_000_rgb.png
│       ├── step_001_rgb.png
│       └── ...
└── ...

Replay Data JSON¶

{
  "episode_id": "001",
  "scene_id": "scene_001",
  "start_position": [0.0, 0.0, 0.0],
  "start_rotation": [1.0, 0.0, 0.0, 0.0],
  "goal_position": [5.0, 0.0, 0.0],
  "trajectory": [
    {
      "step": 0,
      "position": [0.0, 0.0, 0.0],
      "rotation": [1.0, 0.0, 0.0, 0.0],
      "action": {"x": 0.5, "y": 0.0, "yaw": 0.0}
    }
  ],
  "observations": [
    {
      "step": 0,
      "rgb": {
        "face": "base64_encoded_image",
        "left": "base64_encoded_image",
        "right": "base64_encoded_image"
      }
    }
  ],
  "metadata": {
    "success": true,
    "path_length": 4.2,
    "geodesic_distance": 3.0,
    "num_steps": 42
  }
}

Using the Replay Tool¶

Command-Line Tool¶

Single Episode Replay¶

python scripts/replay_eval.py \
    --episode 001 \
    --results eval_results/ \
    --output replay.mp4

Batch Replay¶

python scripts/replay_eval.py \
    --results eval_results/ \
    --output replay_videos/

Custom Parameters¶

python scripts/replay_eval.py \
    --episode 001 \
    --results eval_results/ \
    --output replay.mp4 \
    --fps 15 \
    --resolution 1920x1080

Command-Line Arguments¶

Argument	Description	Default
`--episode EPISODE_ID`	Episode ID (single replay)	-
`--results RESULTS_DIR`	Evaluation results directory	`./eval_results`
`--output OUTPUT_PATH`	Output path	`./replay.mp4`
`--fps FPS`	Video frame rate	`15`
`--resolution WIDTHxHEIGHT`	Video resolution	`1920x1080`
`--show-trajectory`	Show trajectory	`true`
`--show-info`	Show info panel	`true`

Python API¶

Basic Usage¶

from navarena_bench.replay import BaseReplayer
from navarena_bench.replay.loader import ReplayLoader

# Load replay data
loader = ReplayLoader("eval_results/episode_001")
replay_data = loader.load()

# Create replayer
replayer = BaseReplayer.init("video", loader)

# Generate replay video
replayer.replay(output_path="replay.mp4")

Custom Replayer¶

from navarena_bench.replay.base import BaseReplayer
from navarena_bench.replay.loader import ReplayLoader

@BaseReplayer.register("my_replayer")
class MyReplayer(BaseReplayer):
    def __init__(self, loader, **kwargs):
        super().__init__(loader, **kwargs)
        # Initialize

    def replay(self, output_path):
        """Generate replay"""
        # Replay logic
        pass

Replay Renderers¶

TrajectoryRenderer¶

Renders the navigation trajectory on the map.

from navarena_bench.replay.renderers.trajectory_renderer import TrajectoryRenderer

renderer = TrajectoryRenderer(
    pgm_map_path="scene_001_transformed.pgm",
    yaml_config_path="scene_001_transformed.yaml"
)

image = renderer.render(
    trajectory=trajectory,
    start_position=start_pos,
    goal_position=goal_pos
)

ObservationRenderer¶

Renders observation images.

from navarena_bench.replay.renderers.observation_renderer import ObservationRenderer

renderer = ObservationRenderer()

image = renderer.render(
    observations=observations,
    camera_names=["face", "left", "right"]
)

InfoPanelRenderer¶

Renders the evaluation info panel.

from navarena_bench.replay.renderers.info_panel_renderer import InfoPanelRenderer

renderer = InfoPanelRenderer()

image = renderer.render(
    metadata={
        "success": True,
        "path_length": 4.2,
        "geodesic_distance": 3.0,
        "num_steps": 42
    }
)

Replay Video Layout¶

The replay video includes:

Trajectory View - Top left: map and trajectory
Observation View - Top right: multi-camera observations
Info Panel - Bottom: evaluation metrics

Video Layout¶

┌─────────────────┬─────────────────┐
│                 │                 │
│  Trajectory     │  Observations   │
│  (Map+Traj)     │  (Multi-cam)    │
│                 │                 │
├─────────────────┴─────────────────┤
│           Info Panel               │
│   Success: ✓  Path Length: 4.2m   │
└─────────────────────────────────────┘

Replay Configuration¶

replay:
  fps: 15
  resolution: [1920, 1080]
  show_trajectory: true
  show_observations: true
  show_info_panel: true
  trajectory_style:
    line_color: [255, 0, 0]
    line_width: 2
    start_marker: "circle"
    goal_marker: "star"

FAQ¶

Replay data not found

Ensure trajectory saving was enabled during evaluation (save_trajectories: true).

Video generation failed

Check output path permissions and disk space.

Trajectory display incorrect

Verify occupancy grid map file exists and coordinate transform is correct.

Observation images missing

Ensure observation images were saved during evaluation and paths are correct.

See also: Extending · Evaluator Module · Environment Module