Project: DeepBlue Juggling #5 - Ideas for Jugglebot's Eyes
ฝัง
- เผยแพร่เมื่อ 13 มิ.ย. 2024
- In this quick video I cover what my plans are for Jugglebot's eyes. My three current ideas are:
1 - Depth sensing cameras
2 - VR Trackers
3 - Dedicated Motion Tracking System
Links to things from the video:
Depth Cameras: www.kickstarter.com/projects/...
VR Trackers (Tundra Trackers): tundra-labs.com/
VR Accuracy Paper: www.mdpi.com/1424-8220/21/5/1...
OptiTrack System: optitrack.com/systems/#roboti...
00:00 - Intro
01:10 - Idea #1 - Depth Sensing Cameras
02:17 - Idea #2 - VR Trackers
03:46 - Idea #3 - Motion Tracking System
04:25 - Final Remarks - วิทยาศาสตร์และเทคโนโลยี
Do you need eyes? Jugglers can juggle with their eyes closed - could you measure the force used on each leg of the arm to determine where the ball should be going, and move to its expected landing zone?
Hope it's not too windy though!
Haha yep, _some_ jugglers can juggle with their eyes closed, though---even with 10 years of experience---I've never been able to get more than ~10-15 catches of 3 balls when I've tried it myself - it's not easy!
What you're suggesting would absolutely work in theory, but I think putting it into practice would be quite challenging. The problem of wind that you've identified would absolutely be a show-stopper if jugglebot were (visually) blind.
@@harrisonlow That's probably where you can use the Kalman filter: to combine the location prediction from the model (from the forces) with the measurement (from the camera). Both have errors but are more accurate when combined.
@@Ray-jy9sf I agree that jugglebot would be more accurate _in theory_ if it had force sensors in the hand and was able to combine those readings with those from the eyes, though I wonder how well that would translate to real-world accuracy gains. I say this for two reasons:
1) The camera-only approach _should_ be quite accurate if the balls are being thrown any appreciable height. I say this because projectile motion is a very well understood process and if the "eyes" are sampling at, say, 100 Hz, I would expect them to be able to average out any errors and give a very precise prediction of where the ball will be at any point in the future (until it changes path).
2) My (somewhat limited) experience with force sensors is that they are quite noisy and I have no idea how I'd mount force sensors to the hand to get a meaningful reading.
There are some specifics here that relate to the "brain" part of jugglebot and how the juggling model works, which I haven't discussed in any videos yet, so there are undoubtedly many questions you have along these lines. I've been wracking my brain for a while now trying to figure out how to best present that content because it is quite detailed and dry... Perhaps I just need to bite the bullet...
I like the suggestion of a Kalman filter! I have never used this approach myself, but I read into them quite a bit a few years ago and they seem very interesting. Let me know if you have any other suggestions or thoughts on this!
Your goal is to deepblue juggling. That is, you want to beat humans at jugggling. I would say that you should only use option one if you wanted to beat human at juggling in the most fair and accurate way. Humans only have eyes and arms and a brain. If you use option 2 or 3, now you are using tech that gives you a much greater advantage at juggling than humans have available. I know it's just one opinion :) But yes if your goal is to completely destroy humans at juggling, then use any means necesary. I was just concerned with people saying that "of course your system is better at juggling, it has better eyes / better system to judge position etc..".
Haha I like this take!