Quick Start
OpenCDA provides benchmark scenarios that researchers can use directly without any modification. We will also be adding new testing scenarios in future releases. Running these scenario testings only need one line of command:
cd ~/OpenCDA
python opencda.py -t scenario_name -v carla_version --apply_ml --record
Parameters explanation:
-t: The name of the tesitng scenario. A python script with the same name should exist inopencda/scenario_testing/to construct the simulation, and a yaml file with the same name should exist inopencda/scenario_testing/config_yaml/to define the simulation parameters.-v --version: Define the CARLA simulator version, default 0.9.11. Right now only 0.9.11 and 0.9.12 are supported. We need this as there are some api changes from 0.9.11 to 0.9.12.--apply_ml(Optional): A flag to indicate whether a deep learning model needs to be loaded. If this flag is set, Pytorch will be imported.--record(Optional): A flag to indicate whether to record this simulation. Check here for more details.
Below we will demonstrate some examples of running the benchmark testings in OpenCDA.
Single Vehicle Test
**Notice: Change -v 0.9.11 to -v 0.9.12 if you are using CARLA 0.9.12.
1. Two-lane highway test
python opencda.py -t single_2lanefree_carla -v 0.9.11
In this scenario , a single Connected Automated Vehicle will be spawned at a 2-lane highway customized map. This
CAV will try to reach the assigned destination with a desired speed of 100km/h and manage to safely interact
with the surrounding traffic flow. The CAV’s localization, planning, and control modules will be activated, and the perception module will be deactivated
by default, thus pytorch is NOT required in this testing .
If you want to activate the perception module, please check Yaml Defining Rules to see details.
Note: The bounding boxes draw on the camera and lidar are retrieved from the server directly and projected to the sensor space
2. Town06 test (Pytorch required)
python opencda.py -t single_town06_carla -v 0.9.11 --apply_ml
The apply_ml flag will import the pytorch library and load Yolov5 model (Thus Pytorch is required) for object detection. Thus, in this
scenario, the perception, localization, planning and control modules will all be activated.
Note: The bounding box draw here comes from the detection model.
3. Town06 Co-simulation test (Pytorch and Sumo required)
python opencda.py -t single_town06_cosim -v 0.9.11 --apply_ml
This scenario applies Sumo to generate the traffic flow instead of using Carla traffic manager.
Yolov5 and simple Lidar fusion are used to detect object 3D bounding box. Therefore, both
Pytorch and Sumo are required to install to run this benchmark test.
Cooperative Driving Test
1. Platooning stability test
python opencda.py -t platoon_stability_2lanefree_carla -v 0.9.11
In this scenario, a platoon with 4 members will be placed at the 2-lane highway map. The platoon leader will dramatically increases and decreases its speed to test whether the members can still keep the desired time gap. In this way, the platoon stability is verified.
2. Cooperative merge and joining a platoon
python opencda.py -t platoon_joining_2lanefree_carla -v 0.9.11
In this scenario, a platoon will drive on the mainline together with a mixed traffic flow. A single CAV will come from the merging lane, communicate with the platoon to cooperatively merge into the mainline, and simultaneously join the platoon.
3. Cooperative merge and joining a platoon under co-simulation (SUMO required)
python opencda.py -t platoon_joining_2lanefree_cosim -v 0.9.11
4. Platoon back-join (Pytorch required)
python opencda.py -t platoon_joining_town06_carla -v 0.9.11 --apply_ml
A single CAV will try to overtake several human-driven vehicles to join the platoon from the back.
Pytorch is required for this test since we use yolov5 detetion.
