Hey, i’m a student from Germany and searching for a project to work on. I have some experience in fooling visual object detection/classification AI’s. Would creating a database of adversarial examples for those be a possible project in any way?
Sorry for the response delay ! I completely forget about this subsection …
Personally, I am not fan of work on AI currently… I would really prefer work on our codebase and side project as it will profit more to the community
Rafael, welcome! The idea seems to be a bit too general and not Ardupilot specific enough, especially since no CNNs have been fully integrated with Ardupilot yet. So in short it would make more sense to first implement and test an onboard object detection solution, before looking into robustness and studying/coming up with adversarial examples.
Hi guys, Shaswat Dharaiya here. For a quite sometime now I’ve been working on drones, I’ve worked on my own drone. I’ve used IMU(MPU6050) for my drone for orientation balancing or position control and I realised that it can be used for Non-GPS navigation or INS which can provide an amazing accuracy of just few centimetres. Also I’ve used IMUs for gesture control, so i think the drones we use for ArduPilot can be now gesture controlled based. Please do let me know what you think of these projects.
P.S. I’m new to ArduPilot and I read about few threads and completed initial steps, please let me know how else can i contribute
Here’s my github profile if you’re interested to see through the code: