Hello,
I am feeding the distance sensors which is radars (mmWave) into Ardupilot via MAVlink for obstacle avoidance function and terrain following function. In a nutshell, it is okay, everything works well. I have no problems in the reading radars data by Raspberry and creating the MAVlink message to send those information to Ardupilot. My bigger question is about how the Ardupilot processes that signal. To be honest, I do not have time (and a little difficult) to dive into the code for ObstacleAvoidance function but to my understanding, they are built based on Dijkstra for finding the optimal ways out of the obstacles (creating a fence). My questions are:
- For commercial “plug-and-play” distance sensor feeding directly to the hardware (Cube, Pixhawk…), does the Ardupilot apply the filters (lowpass or something) for these or just directly use them? To my observation in the Mission Planer (Crtl + F), the distance suddenly changes. Is it necessary to me for implementing myself the filters in my companion PC?
- Radar sensor returning objects is a little “noisy” because the vertical beam of them will cover the ground if drones are at low attitude, especially for forward direction (if the radar is mounted in front of the drone). Do anyone have experience on this which you could share?
- Any ideas about machine learning methods are extremely welcome.
I will be very happy to get your feed back. Thank you in advance.