Overview

The AWS DeepRacer vehicle is a 1/18th scale RC car that can drive itself using a reinforcement learning model. The vehicle comes with six sample projects and open-source device software, providing developers with the opportunity to prototype robotics applications for use with AWS DeepRacer hardware.

The AWS DeepRacer comes with all required libraries and packages preinstalled. The vehicle's device software includes Ubuntu 20.04 Focal Fossa, Intel® OpenVINO™ toolkit 2021.1.110, ROS 2 Foxy Fitzroy, and Python 3.8.

The DeepRacer Launcher ROS package launches all required AWS DeepRacer core application nodes, including those for perception, decision, navigation, and application support. For more information about the AWS DeepRacer core application and components, refer to information here.

License

The source code is released under Apache 2.0.

Supported ROS Distributions

• Foxy

Software

The AWS DeepRacer device with the upgraded OpenSource software contains the following:

1. ROS 2 Foxy and dependencies: ROS 2 Foxy (ros-foxy-desktop), CV Bridge (ros-foxy-cv-bridge), image transport (ros-foxy-image-transport), standard messages (ros-foxy-std-msgs), and sensor messages (ros-foxy-sensor-msgs) are installed on the DeepRacer device.

2. Intel OpenVino toolkit and its dependencies: Intel OpenVino toolkit (intel-openvino-dev-ubuntu20-2021.1.110) and its dependencies are preinstalled on the AWS DeepRacer device. The AWS DeepRacer core application uses Intel OpenVino to optimize reinforcement learning models and run inference on them.

3. Tensorflow: A custom built Tensorflow v2.4.0 wheel with no AVX support is installed on the AWS DeepRacer device.

4. AWS DeepRacer Debian packages:

   1. aws-deepracer-sample-models: This package installs three sample models deployed by default on the AWS DeepRacer device.

   2. aws-deepracer-util: This package installs utility scripts required by the AWS DeepRacer core application.

   3. aws-deepracer-device-console: This package installs the UI component of the AWS DeepRacer device console.

   4. aws-deepracer-core: This package installs all ROS 2 Foxy packages required for AWS DeepRacer core application functionality.

AWS DeepRacer Core Components

The AWS DeepRacer application includes packages for perception, decision, navigation, and application support.

Perception

Use these nodes to enable the AWS DeepRacer device's use of sensors for perceiving its environment.

• camera_node: This node reads and publishes data from the camera.

• sensor_fusion_node: This node combines and publishes camera and LiDAR data (if a LiDAR sensor is installed) as a single message.

Decision

Use this node to determine the actions the AWS DeepRacer device should perform based on the data it perceives.

• inference_node: This node runs inference on the perceived data using the selected machine learning model.

Navigation

Use these nodes to enable the vehicle's use of its motor and hardware to move through the environment.

• deepracer_navigation_node: This node collects model inference results and publishes a servo message with throttle and steering angle values based on the action space for the selected machine learning model.

• servo_node: This node maps the input servo throttle and servo angle ratios to raw PWM values that set on the servo or motor to move the vehicle.

Application Support

Use these nodes to support core application functionality.

• ctrl_node: As the main node, this exposes services used by backend webserver API calls.

• software_update_node: This node is responsible for the software update system managing the following packages: aws-deepracer-core, aws-deepracer-util, aws-deepracer-sample-models, and aws-deepracer-webserver.

• model_loader_node: This node is responsible for extracting tar.gz model files from USB and those uploaded from the console.

• otg_control_node: This node is responsible for enabling and disabling the OTG connection and publishing the connection status.

• network_monitor_node: This node is responsible for connecting to WiFi based on the configuration file on USB and communicating the network connection status.

• device_info_node: This node provides hardware and software version information.

• battery_node: This node reads and publishes vehicle battery level information.

• model_optimizer_node: This node runs the model optimizer on the selected machine learning model.

• status_led_node: This node controls blinking and solid light functionality for the status LEDs found on the side of the AWS DeepRacer device.

• usb_monitor_node: This node monitors USB connections and publishes a notification if a required file is found.

• webserver_publisher_node: This node is a collection of FLASK APIs called from the front end. These APIs call ROS services and return results to the front end.