### Install DiffusionDriveV2 Packages

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/install.md

Installs the required Python packages for DiffusionDriveV2 using pip within the NAVSIM conda environment.

```bash
conda activate navsim
pip install diffusers einops 
```

--------------------------------

### Initialize SceneLoader and SceneFilter

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Sets up the environment for loading NAVSIM scenes and sensor data. Requires specifying the data split and filter configuration.

```python
import os
from pathlib import Path

import hydra
from hydra.utils import instantiate
import numpy as np
import matplotlib.pyplot as plt

from navsim.common.dataloader import SceneLoader
from navsim.common.dataclasses import SceneFilter, SensorConfig

SPLIT = "mini"  # ["mini", "test", "trainval"]
FILTER = "all_scenes"

hydra.initialize(config_path="../navsim/planning/script/config/common/scene_filter")
cfg = hydra.compose(config_name=FILTER)
scene_filter: SceneFilter = instantiate(cfg)
openscene_data_root = Path(os.getenv("OPENSCENE_DATA_ROOT"))

scene_loader = SceneLoader(
    openscene_data_root / f"navsim_logs/{SPLIT}",
    openscene_data_root / f"sensor_blobs/{SPLIT}",
    scene_filter,
    sensor_config=SensorConfig.build_all_sensors(),
)
```

--------------------------------

### Training DiffusionDrive with Reinforcement Learning

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Initiates the training process for the DiffusionDrive RL agent. Ensure the agent checkpoint path and other parameters are correctly set.

```bash
python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_training.py \
        agent=diffusiondrivev2_rl_agent \
        experiment_name=training_diffusiondrivev2_rl_agent \
        train_test_split=navtrain \
        agent.checkpoint_path=ckpts/diffusiondrive_navsim_88p1_PDMS \
        split=trainval \
        trainer.params.max_epochs=10 \
        cache_path="${NAVSIM_EXP_ROOT}/training_cache/" \
        use_cache_without_dataset=True \
        force_cache_computation=False
```

--------------------------------

### Plot camera frames with annotations

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Visualizes bounding-box annotations within the camera images for a specific frame. Requires `frame_idx`.

```python
from navsim.visualization.plots import plot_cameras_frame_with_annotations

fig, ax = plot_cameras_frame_with_annotations(scene, frame_idx)
plt.show()
```

--------------------------------

### Cache Dataset for Fast Evaluation (Optional)

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Caches the dataset for faster evaluation. This is an optional step for performance optimization.

```bash
python navsim/planning/script/run_dataset_caching.py agent=diffusiondrivev2_rl_agent experiment_name=diffusiondrivev2_cache train_test_split=navtest cache_path=$NAVSIM_EXP_ROOT/metric_feature_cache
```

--------------------------------

### Fast Evaluation of Trained Model

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Evaluates the trained DiffusionDriveV2 model rapidly using a fast PDM score script. Ensure all checkpoint and cache paths are correctly configured.

```bash
python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_pdm_score_fast.py \
        agent=diffusiondrivev2_sel_agent \
        experiment_name=diffusiondrivev2_agent_eval \
        train_test_split=navtest \
        agent.checkpoint_path=ckpts/diffusiondrivev2_sel.ckpt \
        +metric_cache_path="${NAVSIM_EXP_ROOT}/metric_cache/" \
        +test_cache_path="${NAVSIM_EXP_ROOT}/metric_feature_cache/"
```

--------------------------------

### Accelerated Data Download

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/install.md

Uses a provided script to parallelize data downloads, enabling faster acquisition of necessary files.

```bash
cd /path/to/DiffusionDriveV2/download
bash super_download.sh
```

--------------------------------

### Cache Dataset for Training

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Caches the dataset for training purposes. Ensure the agent and experiment names are correctly specified.

```bash
python navsim/planning/script/run_dataset_caching.py agent=diffusiondrivev2_rl_agent experiment_name=diffusiondrivev2_cache train_test_split=navtrain
```

--------------------------------

### Load a random scene

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Selects a random scene token from the loaded scenes and retrieves the corresponding scene data.

```python
token = np.random.choice(scene_loader.tokens)
scene = scene_loader.get_scene_from_token(token)
```

--------------------------------

### Cache Dataset for Evaluation

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Caches the dataset for evaluation. The cache path should point to the metric cache directory.

```bash
python navsim/planning/script/run_metric_caching.py train_test_split=navtest cache.cache_path=$NAVSIM_EXP_ROOT/metric_cache
```

--------------------------------

### Download GTRS Trajectories

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/install.md

Downloads simulated ground-truth trajectories for GTRS vocabulary, used for data augmentation in mode selector training.

```bash
cd gtrs_traj
wget https://huggingface.co/Zzxxxxxxxx/gtrs/resolve/main/navtrain_16384.pkl
```

--------------------------------

### Generate GIF from Scene Frames

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Use `frame_plot_to_gif` to create an animated GIF from a scene's frames. This function requires a file name for the output GIF, a plotting function (e.g., `plot_cameras_frame_with_annotations`), the scene object, and a list of frame indices to include.

```python
from navsim.visualization.plots import frame_plot_to_gif

frame_indices = [idx for idx in range(len(scene.frames))]  # all frames in scene
file_name = f"./{token}.gif"
images = frame_plot_to_gif(file_name, plot_cameras_frame_with_annotations, scene, frame_indices)
```

--------------------------------

### Training Mode Selector

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Trains the mode selector agent. This script requires the RL-trained DiffusionDrive model checkpoint.

```bash
python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_training.py \
        agent=diffusiondrivev2_sel_agent \
        experiment_name=training_diffusiondrivev2_sel_agent \
        train_test_split=navtrain \
        agent.checkpoint_path=ckpts/diffusiondrivev2_rl.ckpt \
        split=trainval \
        trainer.params.max_epochs=20 \
        cache_path="${NAVSIM_EXP_ROOT}/training_cache/" \
        use_cache_without_dataset=True \
        force_cache_computation=False
```

--------------------------------

### Plot camera frames with LiDAR

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Overlays the LiDAR point cloud onto the camera images for a given frame. Requires `frame_idx`.

```python
from navsim.visualization.plots import plot_cameras_frame_with_lidar

fig, ax = plot_cameras_frame_with_lidar(scene, frame_idx)
plt.show()
```

--------------------------------

### Plot BEV with agent trajectory

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Visualizes the Birds-Eye-View of the scene along with the predicted trajectory of an agent compared to the human driver. Requires an instantiated agent object.

```python
from navsim.visualization.plots import plot_bev_with_agent
from navsim.agents.constant_velocity_agent import ConstantVelocityAgent

agent = ConstantVelocityAgent()
fig, ax = plot_bev_with_agent(scene, agent)
plt.show()
```

--------------------------------

### Plot camera frames

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Displays all eight surrounding camera views in a 3x3 grid, with the BEV plot in the center. Uses `frame_idx` to select the specific frame.

```python
from navsim.visualization.plots import plot_cameras_frame

fig, ax = plot_cameras_frame(scene, frame_idx)
plt.show()
```

--------------------------------

### Cache Dataset for PDMS Calculation

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Caches the dataset for calculating PDMS during training. Specify the appropriate train test split and cache path.

```bash
python navsim/planning/script/run_metric_caching.py train_test_split=navtrain cache.cache_path=$NAVSIM_EXP_ROOT/train_pdm_cache
```

--------------------------------

### Create custom BEV plot with annotations and LiDAR

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Constructs a custom Birds-Eye-View plot by manually adding annotations and LiDAR data to a matplotlib Axes object. Utilizes helper functions for configuration.

```python
from navsim.visualization.plots import configure_bev_ax
from navsim.visualization.bev import add_annotations_to_bev_ax, add_lidar_to_bev_ax


fig, ax = plt.subplots(1, 1, figsize=(6, 6))

ax.set_title("Custom plot")

add_annotations_to_bev_ax(ax, scene.frames[frame_idx].annotations)
add_lidar_to_bev_ax(ax, scene.frames[frame_idx].lidar)

# configures frame to BEV view
configure_bev_ax(ax)

plt.show()
```

--------------------------------

### Official Evaluation Script

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/docs/train_eval.md

Uses the official evaluation script for evaluating the trained model. This may require uncommenting a specific line in the model's Python file.

```bash
python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_pdm_score.py \
        train_test_split=navtest \
        agent=diffusiondrivev2_sel_agent \
        worker=ray_distributed \
        agent.checkpoint_path=ckpts/diffusiondrivev2_sel.ckpt \
        experiment_name=diffusiondrivev2_agent_eval
```

--------------------------------

### Plot Birds-Eye-View (BEV) frame

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/tutorial/tutorial_visualization.ipynb

Visualizes a single frame of the scene in a Birds-Eye-View. Use `frame_idx` to specify which historical or future frame to display.

```python
from navsim.visualization.plots import plot_bev_frame

frame_idx = scene.scene_metadata.num_history_frames - 1 # current frame
fig, ax = plot_bev_frame(scene, frame_idx)
plt.show()
```

--------------------------------

### DiffusionDriveV2 BibTeX Citation

Source: https://github.com/hustvl/diffusiondrivev2/blob/master/README.md

Use this BibTeX entry to cite DiffusionDriveV2 in academic publications. Ensure your LaTeX environment supports BibTeX.

```bibtex
@article{zou2025diffusiondrivev2,
  title={DiffusionDriveV2: Reinforcement Learning-Constrained Truncated Diffusion Modeling in End-to-End Autonomous Driving},
  author={Zou, Jialv and Chen, Shaoyu and Liao, Bencheng and Zheng, Zhiyu and Song, Yuehao and Zhang, Lefei and Zhang, Qian and Liu, Wenyu and Wang, Xinggang},
  journal={arXiv preprint arXiv:2512.07745},
  year={2025}
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.