Overview

File tree:

.
├── compile_proto.sh            # script for compiling proto files
├── cyber                       # proto files from Apollo Autopilot
├── cyber_bridge                # codes for communicating with Apollo cyber_bridge
├── Dockerfile                  # Dockerfile
├── dreamview_api.py            # Apollo Dreamview websocket api
├── main.py                     # main script
├── modules                     # proto files from Apollo Autopilot
├── opendrive                   # opendrive maps
├── pygame_viewer.py            # pygame viewer sub-process
├── README.md                   # this README.md
├── requirements.txt            # required python packages
├── scenario_configs            # scenario configurations, unused for now
├── scenario_runner             # scenario_runner submodule
├── sensor_configs              # sensors configurations
├── sensors                     # codes for sensors
├── setup.bash                  # setup environment variables
└── utils.py                    # common utilities codes

How to run

# 0. good programmers start counting at zero
git clone --recurse-submodules git@gitlab.quyan.info:the-flying-hoyshanist/simulator.git
# 1. cd to project root
cd simulator/
# 2. generate protobuf files
bash compile_proto.sh
# 3. source setup script
source setup.bash
# 4. show usage
python main.py -h
# 5. run main.py script
python main.py --carla-host <carla.server.ip.addr> --apollo-host <apollo.ip.addr> --configFile scenario_configs/781_stop_at_fix_location_cfg.json

TO-DO

1. include scenario_runner as a git submodule, and start scenerio_manager as a child process
2. turn all prints into logs
3. project docker image (Dockerfile)
4. apollo opendrive map converter (*.xodr -> apollo_map)
5. communication between scenario_runner and main process
6. different frequency for different sensors
7. sensor configuration files

Components

A functioning simulator node should contains the following components:

• Carla Sensors Module
• Control Sensor Module
• Pygame Viewer
• Scenario Runner
• Apollo Bootstrap

Carla Sensors Module

Algorithm:

1. Wait for ego vehicle to be created
2. Setup sensors on ego
3. For every server tick, update sensors and send msg to Apollo
4. When done, destroy all created sensors (Carla actors)

Exit condition: Process exits when ego no longer exists

Control Sensor Module

Algorithm:

1. Wait for ego vehicle to be created
2. For every server tick, receive apollo control commands and apply to ego

Exit condition: Process exits when ego no longer exists

Pygame Viewer

Algorithm:

1. Wait for ego vehicle to be created
2. Create an rgb camera sensor attached to ego, and pipe output dataframe to pygame
3. For every server tick, update ego status and pygame windows
4. When done, destroy all created sensors (Carla actors)

Exit condition: Process exits when one of the following happens:

1. Ego vehicle no longer exists
2. User prompted exit (via pygame interface)

Scenario Runner (TODO)

Algorithm:

1. Setup Carla world (map, weather, etc)
2. Create ego vehicle and other actors (vehicles and walkers)
3. For every server tick, monitor ego vehicle metrics and manipulate other actors' behavour accordingly
4. Output scenario report
5. Destroy all carla actors created in step 2

Exit condition: Process exits when one of the following happens:

1. Timeout
2. Ego vehicle violate one of the criteria

Apollo Bootstrap (TODO)

Algorithm:

1. Setup Apollo map
2. Setup Apollo vehicle
3. Load all required Apollo modules
4. Send Apollo routing request
5. Monitor Apollo moduels status
6. Output abnormalies if any

Exit condition: Process exits when indicated (probably a multiprocessing.Event?)

Available Sensors

1. ChassisAlter
2. BestPose
3. Odometry
4. InsStatus
5. CorrectedImu
6. Obstacles
7. TrafficLightAlter
8. ClockSensor

Common attributes for all sensors:

• "name": string, id of the sensor, used for identification, should be different for different sensor instances
• "type": string, type of the sensor class
• "frequency": number, set it to -1.0 if the sensor should be run as fast as it can

ChassisAlter

This sensor sends data about the vehicle chassis.

BestPose

This sensor outputs the GPS location of the vehicle in Longitude/Latitude and Northing/Easting coordintates.

Odometry

This sensor outputs the GPS location of the vehicle in Longitude/Latitude and Northing/Easting coordintates and the vehicle velocity.

Since some maps do not centred at (0,0), there are two additional attributes for this sensor:

• "x_offset": number, offset along the x axis
• "y_offset": number, offset along the y axis

InsStatus

This sensor outputs the status of the GPS correction due to INS. The simulator is an ideal environment in which GPS is always corrected.

CorrectedImu

This sensor output the data about the vehicle imu status.

Obstacles

This sensor returns 3D ground truth data for training and creates bounding boxes around the detected objects.

Additional attributes:

• "x_offset": number, offset along the x axis
• "y_offset": number, offset along the y axis
• "vehicle_distance": number, vehicles outside of this radius will not be detected
• "walker_distance": number, pedestrians outside of this radius will not be detected

TrafficLightAlter

This sensor returns ground truth data for traffic light signals connected to the current lane of ego vehicle.

Clock

This sensor outputs simulated time to CyberRT as clock messages.

Using Docker

Build docker image

git clone --recurse-submodules git@gitlab.quyan.info:the-flying-hoyshanist/simulator.git
cd ./simulator
docker build -t simulator-0.9.11:v0.1 .

Start docker container

docker run -it -d \
-v /tmp/.X11-unix/:/tmp/.X11-unix \
simulator-0.9.11:v0.1 \
bash

Misc

Compile protos

bash compile_proto.sh

About Apollo Map

• Apollo Autopilot maps use a modified version of opendrive, so directly export carla maps to opendrive format won't be compatible with Apollo.
• AuroAI has converted all carla maps to Apollo maps format, see here

About lgsvl Maps

lgsvl/simulator#1416

Disable carla server rendering for higher fps

# in carla/PythonAPI/util
python3 config.py --no-rendering

See https://carla.readthedocs.io/en/0.9.11/adv_rendering_options/ for details.