Intro

Team Soulless have competed in the Self Racing Cars event, and the code in the repository is the autonomy stack which ran on a Kia Niro with a drive by wire DriveKit from Polysync.

Overview

The code is split into different rosnodes with separate functionalities:

1. Drive by wire / Joysticking the vehicle:

Our team forked Polysync's ROSCCO package which has a simple joystick controller that we have tested with a PS4 controller, and compataible with an XBOX controller. We have only added an additional node to test getting reported values from the vehicle to make sure our canbus is setup properly.

The node we have added is in the roscco_report.cpp file.

2. Perception:

SegNet was used for semantic segmentation, however with the twist of using YOLOv2 for feature extraction. Thunderhill's Semantic segmentation is availble at: https://github.com/jendrikjoe/SegNet/tree/thunderhill2019

3. Trajector Generation:

Using the semantic segmentation processed image, we have looked at the driveable regions of the road using lane markings and cones as boundaries, our algorithm drew trajector lines as the example shown below:

More examples of painted trajectories are under the traject_thunderhill/ repository.

Details of the algorithm have been implemented in iterations, and a python notebook is availble to view here

4. Velocity PID controller and OBD reader

We are using ROS's PID package with values tuned in the this launch file. The throttle and steering values of the vehicles are read through two simple nodes, that look as follows:

CAN_SPEED_ID = 0x52A
KIA_SOUL_OBD_WHEEL_SPEED_CAN_ID = 0x386
KIA_SOUL_OBD_STEERING_WHEEL_ANGLE_CAN_ID = 0x2B0
KIA_SOUL_OBD_STEERING_ANGLE_SCALAR = 0.1

...

if can_id == KIA_SOUL_OBD_WHEEL_SPEED_CAN_ID:
             if not frame.data:
                  return
             frame_data = [struct.unpack('B', x)[0] for x in frame.data]
             speed = 0
             for i in range(4):
                 raw = (frame_data[i*2+1] & 0x0F) << 8 | frame_data[i*2]
                 wheel_speed = int(raw / 3.2) / 10.0;
                 speed += wheel_speed
             speed /= 4
             msg = Float64()
             msg.data = speed
             self.speed_pub.publish(msg)

Availble at: obd_reader.py

Running the code:

In order to get all the dependencies there is an install.sh file providing you with all you need to install (Might need some additions, feel free to open an issue if you get stuck).

To check all related code for the submodules, run gitall.sh, and to make the project run make.sh

Launching the code:

There is a launch.sh file to provide a great entry point to the project. However all launch files are also availble under: https://github.com/ranakhalil/Soulless2019/tree/master/src/launch

Have any questions/suggestion/comments?

Feel free to open an issue or a PR to our repo. We look forward to hearing from you.