Rebuttal (#15)

* pg+nuplan train

* Need map

* use gym wrapper

* use createGymWrapper

* doc

* use all scenarios!

* update 80000 scenario

* train script

documentation/Makefile

@@ -0,0 +1,20 @@
+# Minimal makefile for Sphinx documentation
+#
+
+# You can set these variables from the command line, and also
+# from the environment for the first two.
+SPHINXOPTS    ?=
+SPHINXBUILD   ?= sphinx-build
+SOURCEDIR     = source
+BUILDDIR      = build
+
+# Put it first so that "make" without argument is like "make help".
+help:
+	@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+.PHONY: help Makefile
+
+# Catch-all target: route all unknown targets to Sphinx using the new
+# "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
+%: Makefile
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)

documentation/README.md

@@ -0,0 +1,9 @@
+This folder contains files for the documentation: [https://scenarionet.readthedocs.io/](https://scenarionet.readthedocs.io/).
+
+To build documents locally, please run the following codes:
+
+```
+pip install sphinx sphinx_rtd_theme
+cd scenarionet/documentation
+make html
+```

documentation/make.bat

@@ -0,0 +1,35 @@
+@ECHO OFF
+
+pushd %~dp0
+
+REM Command file for Sphinx documentation
+
+if "%SPHINXBUILD%" == "" (
+	set SPHINXBUILD=sphinx-build
+)
+set SOURCEDIR=source
+set BUILDDIR=build
+
+if "%1" == "" goto help
+
+%SPHINXBUILD% >NUL 2>NUL
+if errorlevel 9009 (
+	echo.
+	echo.The 'sphinx-build' command was not found. Make sure you have Sphinx
+	echo.installed, then set the SPHINXBUILD environment variable to point
+	echo.to the full path of the 'sphinx-build' executable. Alternatively you
+	echo.may add the Sphinx directory to PATH.
+	echo.
+	echo.If you don't have Sphinx installed, grab it from
+	echo.http://sphinx-doc.org/
+	exit /b 1
+)
+
+%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+goto end
+
+:help
+%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+
+:end
+popd

documentation/source/conf.py

@@ -0,0 +1,55 @@
+# Configuration file for the Sphinx documentation builder.
+#
+# This file only contains a selection of the most common options. For a full
+# list see the documentation:
+# https://www.sphinx-doc.org/en/master/usage/configuration.html
+
+# -- Path setup --------------------------------------------------------------
+
+# If extensions (or modules to documentation with autodoc) are in another directory,
+# add these directories to sys.path here. If the directory is relative to the
+# documentation root, use os.path.abspath to make it absolute, like shown here.
+#
+# import os
+# import sys
+# sys.path.insert(0, os.path.abspath('.'))
+
+
+# -- Project information -----------------------------------------------------
+
+project = 'ScenarioNet'
+copyright = 'MetaDriverse'
+author = 'MetaDriverse'
+
+# The full version, including alpha/beta/rc tags
+release = '0.1.1'
+
+# -- General configuration ---------------------------------------------------
+
+# Add any Sphinx extension module names here, as strings. They can be
+# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
+# ones.
+extensions = [
+    "sphinx.ext.autosectionlabel",
+    "sphinx_rtd_theme"
+]
+
+# Add any paths that contain templates here, relative to this directory.
+templates_path = ['_templates']
+
+# List of patterns, relative to source directory, that match files and
+# directories to ignore when looking for source files.
+# This pattern also affects html_static_path and html_extra_path.
+exclude_patterns = []
+
+# -- Options for HTML output -------------------------------------------------
+
+# The theme to use for HTML and HTML Help pages.  See the documentation for
+# a list of builtin themes.
+#
+html_theme = 'sphinx_rtd_theme'
+
+# Add any paths that contain custom static files (such as style sheets) here,
+# relative to this directory. They are copied after the builtin static files,
+# so a file named "default.css" will overwrite the builtin "default.css".
+html_static_path = []

documentation/source/index.rst

@@ -0,0 +1,33 @@
+########################
+ScenarioNet Documentation
+########################
+
+
+Welcome to the ScenarioNet documentation!
+ScenarioNet is an open-sourced platform for large-scale traffic scenario modeling and simulation with the following features:
+
+* ScenarioNet defines a unified scenario description format containing HD maps and detailed object annotations.
+* ScenarioNet provides tools to build and manage databases built from various data sources including real-world datasets like Waymo, nuScenes, Lyft L5, and nuPlan datasets and synthetic datasets like the procedural generated ones and safety-critical ones.
+* Scenarios recorded in this format can be replayed in the digital twins with multiple views, ranging from Bird-Eye-View layout to realistic 3D rendering.
+
+It can thus support several applications including large-scale scenario generation, AD testing, imitation learning, and reinforcement learning in both single-agent and multi-agent settings. The results imply scaling up the training data brings new research opportunities in machine learning and autonomous driving.
+
+This documentation brings you the information on installation, usages and more of ScenarioNet!
+You can also visit the `GitHub repo <https://github.com/metadriverse/scenarionet>`_ and `Webpage <https://metadriverse.github.io/scenarionet/>`_ for code and videos.
+Please feel free to contact us if you have any suggestions or ideas!
+
+
+Citation
+########
+
+You can read `our white paper <https://arxiv.org/pdf/2306.12241.pdf>`_ describing the details of ScenarioNet! If you use ScenarioNet in your own work, please cite:
+
+.. code-block:: latex
+
+    @article{li2023scenarionet,
+      title={ScenarioNet: Open-Source Platform for Large-Scale Traffic Scenario Simulation and Modeling},
+      author={Li, Quanyi and Peng, Zhenghao and Feng, Lan and Duan, Chenda and Mo, Wenjie and Zhou, Bolei and others},
+      journal={arXiv preprint arXiv:2306.12241},
+      year={2023}
+    }
+

scenarionet/tests/script/run_ckpt.py

@@ -1,16 +1,13 @@
 import pygame
 from metadrive.engine.asset_loader import AssetLoader
 from metadrive.envs.real_data_envs.nuscenes_env import ScenarioEnv
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 from scenarionet_training.train_utils.utils import initialize_ray, get_function
 from scenarionet_training.scripts.train_nuplan import config
 
 if __name__ == "__main__":
     initialize_ray(test_mode=False, num_gpus=1)
-    env = GymEnvWrapper(
+    env = createGymWrapper(ScenarioEnv)({
-        dict(
-            inner_class=ScenarioEnv,
-            inner_config={
                 # "data_directory": AssetLoader.file_path("nuscenes", return_raw_style=False),
                 "data_directory": "D:\\scenarionet_testset\\nuplan_test\\nuplan_test_w_raw",
                 "use_render": True,
@@ -44,7 +41,7 @@ if __name__ == "__main__":
                 ),
             }
         )
-    )
+
 
     # env.reset()
     #

scenarionet_training/scripts/local_test.py

@@ -1,7 +1,7 @@
 import os.path
 
 from metadrive.envs.scenario_env import ScenarioEnv
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 
 from scenarionet import SCENARIONET_REPO_PATH, SCENARIONET_DATASET_PATH
 from scenarionet_training.train_utils.multi_worker_PPO import MultiWorkerPPO
@@ -14,8 +14,8 @@ if __name__ == '__main__':
     stop = int(100_000_000)
 
     config = dict(
-        env=GymEnvWrapper,
+        env=createGymWrapper(ScenarioEnv),
-        env_config=dict(inner_class=ScenarioEnv, inner_config=dict(
+        env_config=dict(
             # scenario
             start_scenario_index=0,
             num_scenarios=32,
@@ -34,7 +34,7 @@ if __name__ == '__main__':
 
             # training
             horizon=None,
-        )),
+        ),
 
         # # ===== Evaluation =====
         evaluation_interval=2,

scenarionet_training/scripts/train_nuplan.py

@@ -1,13 +1,13 @@
 import os.path
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 from metadrive.envs.scenario_env import ScenarioEnv
 from scenarionet import SCENARIONET_REPO_PATH, SCENARIONET_DATASET_PATH
 from scenarionet_training.train_utils.multi_worker_PPO import MultiWorkerPPO
 from scenarionet_training.train_utils.utils import train, get_train_parser, get_exp_name
 
 config = dict(
-    env=GymEnvWrapper,
+    env=createGymWrapper(ScenarioEnv),
-    env_config=dict(inner_class=ScenarioEnv, inner_config=dict(
+    env_config=dict(
         # scenario
         start_scenario_index=0,
         num_scenarios=40000,
@@ -16,7 +16,7 @@ config = dict(
 
         # curriculum training
         curriculum_level=100,
-        target_success_rate=0.8, # or 0.7
+        target_success_rate=0.7,
         # episodes_to_evaluate_curriculum=400,  # default=num_scenarios/curriculum_level
 
         # traffic & light
@@ -42,7 +42,7 @@ config = dict(
 
         vehicle_config=dict(side_detector=dict(num_lasers=0))
 
-    )),
+    ),
 
     # ===== Evaluation =====
     evaluation_interval=15,

scenarionet_training/scripts/train_pg.py

@@ -1,14 +1,14 @@
 import os.path
 from ray.tune import grid_search
 from metadrive.envs.scenario_env import ScenarioEnv
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 from scenarionet import SCENARIONET_REPO_PATH, SCENARIONET_DATASET_PATH
 from scenarionet_training.train_utils.multi_worker_PPO import MultiWorkerPPO
 from scenarionet_training.train_utils.utils import train, get_train_parser, get_exp_name
 
 config = dict(
-    env=GymEnvWrapper,
+    env=createGymWrapper(ScenarioEnv),
-    env_config=dict(inner_class=ScenarioEnv, inner_config=dict(
+    env_config=dict(
         # scenario
         start_scenario_index=0,
         num_scenarios=40000,
@@ -21,7 +21,7 @@ config = dict(
         # episodes_to_evaluate_curriculum=400,  # default=num_scenarios/curriculum_level
 
         # traffic & light
-        reactive_traffic=False,
+        reactive_traffic=True,
         no_static_vehicles=True,
         no_light=True,
         static_traffic_object=True,
@@ -41,7 +41,7 @@ config = dict(
 
         vehicle_config=dict(side_detector=dict(num_lasers=0))
 
-    )),
+    ),
 
     # ===== Evaluation =====
     evaluation_interval=15,

scenarionet_training/scripts/train_pg_nuplan.py

@@ -0,0 +1,99 @@
+import os.path
+from ray import tune
+from metadrive.envs.gym_wrapper import createGymWrapper
+from metadrive.envs.scenario_env import ScenarioEnv
+from scenarionet import SCENARIONET_REPO_PATH, SCENARIONET_DATASET_PATH
+from scenarionet_training.train_utils.multi_worker_PPO import MultiWorkerPPO
+from scenarionet_training.train_utils.utils import train, get_train_parser, get_exp_name
+
+config = dict(
+    env=createGymWrapper(ScenarioEnv),
+    env_config=dict(
+        # scenario
+        start_scenario_index=20000,
+        num_scenarios=40000, # 0-40000 nuplan, 40000-80000 pg
+        data_directory=os.path.join(SCENARIONET_DATASET_PATH, "pg_nuplan_train"),
+        sequential_seed=True,
+        no_map=True,
+        # store_map=False,
+        # store_data=False,
+
+        # curriculum training
+        curriculum_level=100,
+        target_success_rate=0.7,
+        # episodes_to_evaluate_curriculum=400,  # default=num_scenarios/curriculum_level
+
+        # traffic & light
+        reactive_traffic=True,
+        no_static_vehicles=True,
+        no_light=True,
+        static_traffic_object=True,
+
+        # training scheme
+        horizon=None,
+        driving_reward=9,
+        steering_range_penalty=1.0,
+        heading_penalty=1,
+        lateral_penalty=1.0,
+        no_negative_reward=True,
+        on_lane_line_penalty=0,
+        crash_vehicle_penalty=1,
+        crash_human_penalty=1,
+        crash_object_penalty=0.5,
+        # out_of_road_penalty=2,
+        max_lateral_dist=2,
+        # crash_vehicle_done=True,
+
+        vehicle_config=dict(side_detector=dict(num_lasers=0))
+
+    ),
+
+    # ===== Evaluation =====
+    evaluation_interval=15,
+    evaluation_num_episodes=1000,
+    evaluation_config=dict(env_config=dict(start_scenario_index=0,
+                                           num_scenarios=1000,
+                                           sequential_seed=True,
+                                           curriculum_level=1,  # turn off
+                                           data_directory=os.path.join(SCENARIONET_DATASET_PATH, "nuplan_test"))),
+    evaluation_num_workers=10,
+    metrics_smoothing_episodes=10,
+
+    # ===== Training =====
+    model=dict(fcnet_hiddens=[512, 256, 128]),
+    horizon=600,
+    num_sgd_iter=20,
+    lr=1e-4,
+    rollout_fragment_length=500,
+    sgd_minibatch_size=200,
+    train_batch_size=50000,
+    num_gpus=0.5,
+    num_cpus_per_worker=0.3,
+    num_cpus_for_driver=1,
+    num_workers=20,
+    framework="tf"
+)
+
+if __name__ == '__main__':
+    # PG data is generated with seeds 10,000 to 60,000
+    args = get_train_parser().parse_args()
+    exp_name = get_exp_name(args)
+    stop = int(100_000_000)
+    config["num_gpus"] = 0.5 if args.num_gpus != 0 else 0
+
+    train(
+        MultiWorkerPPO,
+        exp_name=exp_name,
+        save_dir=os.path.join(SCENARIONET_REPO_PATH, "experiment"),
+        keep_checkpoints_num=5,
+        stop=stop,
+        config=config,
+        num_gpus=args.num_gpus,
+        # num_seeds=args.num_seeds,
+        num_seeds=4,
+        test_mode=args.test,
+        # local_mode=True,
+        # TODO remove this when we release our code
+        # wandb_key_file="~/wandb_api_key_file.txt",
+        wandb_project="scenarionet",
+    )

scenarionet_training/scripts/train_waymo.py

@@ -1,13 +1,13 @@
 import os.path
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 from metadrive.envs.scenario_env import ScenarioEnv
 from scenarionet import SCENARIONET_REPO_PATH, SCENARIONET_DATASET_PATH
 from scenarionet_training.train_utils.multi_worker_PPO import MultiWorkerPPO
 from scenarionet_training.train_utils.utils import train, get_train_parser, get_exp_name
 
 config = dict(
-    env=GymEnvWrapper,
+    env=createGymWrapper(ScenarioEnv),
-    env_config=dict(inner_class=ScenarioEnv, inner_config=dict(
+    env_config=dict(
         # scenario
         start_scenario_index=0,
         num_scenarios=40000,
@@ -41,7 +41,7 @@ config = dict(
 
         vehicle_config=dict(side_detector=dict(num_lasers=0))
 
-    )),
+    ),
 
                     # ===== Evaluation =====
                     evaluation_interval=15,

scenarionet_training/train_utils/utils.py

@@ -9,7 +9,7 @@ import numpy as np
 import tqdm
 from metadrive.constants import TerminationState
 from metadrive.envs.scenario_env import ScenarioEnv
-from metadrive.envs.gym_wrapper import GymEnvWrapper
+from metadrive.envs.gym_wrapper import createGymWrapper
 from ray import tune
 from ray.tune import CLIReporter
 
@@ -292,7 +292,7 @@ def eval_ckpt(config,
         episodes_to_evaluate_curriculum=num_scenarios,
         data_directory=scenario_data_path,
         use_render=render))
-    env = GymEnvWrapper(dict(inner_class=ScenarioEnv, inner_config=env_config))
+    env = createGymWrapper(ScenarioEnv)(env_config)
 
     super_data = defaultdict(list)
     EPISODE_NUM = env.config["num_scenarios"]