SMARTS supports loading and replaying scenarios from the Waymo Motion Dataset. See the Waymo website for more info and download instructions.


Download the dataset files from the dataset website. It is recommended to download the dataset files from the uncompressed/scenario/training_20s folder as they have the full traffic capture for each scenario. Note: Waymo provides 2 different formats for the dataset files. SMARTS expects the Scenario protos format (not the tf.Example protos format). It is also recommended to use version 1.1 of the dataset, which includes enhanced map information.


SMARTS provides some command-line tools to assist with visualizing and selecting Waymo scenarios.

$ scl waymo --help
Usage: scl waymo [OPTIONS] COMMAND [ARGS]...

Utilities for using the Waymo Motion Dataset with SMARTS. See `scl waymo
COMMAND --help` for further options.

--help  Show this message and exit.

export    Export the Waymo scenario to a SMARTS scenario.
overview  Display summary info for each scenario in the TFRecord file.
preview   Plot the map and trajectories of the scenario.

To see all the scenario IDs in a tfrecord file:

$ scl waymo overview ~/waymo/uncompressed_scenario_training_20s_training_20s.tfrecord-00000-of-01000
Scenario ID         Timestamps    Vehicles    Pedestrians
----------------  ------------  ----------  -------------
c84cde79e51b087c           199         151             37
6cec26a9347e8574           199         165             13
fe6141aeb4061824           198          74              6

To preview a scenario:

$ scl waymo preview ~/waymo/uncompressed_scenario_training_20s_training_20s.tfrecord-00000-of-01000 ef903b7abf6fc0fa

Additionally, there are some other standalone tools in the waymo module

  • gen_sumo_map.py: a command-line program that converts the map from a Waymo scenario to a SUMO map

Example Scenario

An example SMARTS scenario is located here. After downloading the dataset, modify the scenario_id and dataset_path variables to point to the desired Waymo scenario:

dataset_path = "/home/user/waymo/uncompressed_scenario_training_20s_training_20s.tfrecord-00000-of-01000"
scenario_id = "ef903b7abf6fc0fa"

You can then run any of the examples with this scenario:

$ scl run --envision examples/e1_egoless.py scenarios/waymo


SMARTS includes a module for loading Waymo scenarios (located in smarts/waymo/waymo_open_dataset) that should work without any configuration. However, depending on the version of protobuf installed, some issues may come up which require the Python files to be regenerated. See this README for instructions on how to do this.