Woohooo, I'm going to Tokyo!
Our paper was accepted to a robotics conference. We have the pleasure of playing with tiny drones, and we get paid and rewarded for it!

(The girl in the video is my amazing collaborator, Hanna)
https://youtube.com/watch/...

Congrats!

Why is the drone so slow btw? And why does it go up & down?

Congratulations!
Look forward to the papers!

I'm a big fan of your robotics projects and actually jealous even :-)

@retoor Because Hanna is flying it without directly seeing it, just through the images that are sent to the laptop. The wifi module onboard is weak - bandwidth and latency issues.

@NeatNerdPrime

https://arxiv.org/pdf/...

That drone is adorable, congratulations!
Enjoy the food.

I'm assuming you're using opencv for this. Is the feed being fed to a separate machine or onboard? Congrats! Here's your first JP word that you need to practice: SUMIMASEN!

@fruitfcker We are not using OpenCV. The onboard computing unit doesn't even have a real OS. We have our own deployment pipeline for running quantized inference onboard.

Very cool - the speed of the detection mapping is super impressive. I can’t imagine the work that went in. Hope you win an award!

That's amazing! You should be very proud of yourself. I checked out some of your robotic projects and I think it's all really cool! I too, work with robots - though they're not the kind that move around, more the industrial type. :-)

Wow, what an incredible achievement! Getting your paper accepted to a robotics conference in Tokyo is a huge deal. https://fnfgo.org

Congratulations on the paper!! Must've worked real hard on it.

@NickyBones the wifi problems come from the fact that its running on 2.4 GHz and indoor.

Also why is camera Acquisition limited to 50ms(paper didn't repeat it, or was it in the 2022 paper)?

Also a subsample to grayscale would massively reduce datavolume to only 1/4 of the color data.

@max19931 2.4GHz is what the module supports, so we can't get picky here.

We need colors for the object detection.

According to spec the camera should give you 90fps at 640x480, but that's actually not happening. The firmware is trash, and we had to re-implement the debayering in software so that's also slowing us down. A big issue is that a lot of things, including the camera, use the same mDMA and then if you try to acquire the images at high freq you basically block other tasks.

@NickyBones it just occurred to me, is this for the Japan IT Week? I've been there a few times. Lots of salary men. And innovations.

@fruitfcker I think it's just a robotics conference that happens to be in Tokyo this year. Last year it was in London....