gemini-2.0-flash-exp (cost: $0.50) Β· Anfrage: 2025-02-07T15:01:47.310807
Generated by rs-summarizer Version unbekannt
Progress on Juggle Bot's Calibration and Tracking System
0:00-0:01 Meeting with QTM: The presenter had a helpful meeting with someone from QTM (Qualisys), the company that makes the motion capture cameras.
0:01-0:02 Erroneous Markers: The presenter is still getting weird erroneous markers detected by the system.
0:02-0:03 Origin Alignment: The global coordinate system is now aligned with the base's origin after calibration.
0:03:20-0:04:00 Origin Translation and Rotation: QTM allows translating and rotating a rigid body's origin, aligning it with specific markers or geometric centers. It is possible to assign XYZ coordinates by relying on a single known marker.
0:04:40 Rigid Body as Sphere: If a rigid body is a sphere, its origin can be set to the spherical centroid. The presenter wants to improve tracking of the ball by capturing rotation of it. Potentially by covering it in polka dot-like dots.
0:05:30-0:06:56 Ball Trajectory: The presenter and John have noted that the balls don't fly straight, even when thrown by the robot.
0:07:00-0:11:17 Transforming coordinate systems: Explains how to transform the global coordinate system to be the same coordinate system of a certain rigid body (e.g., base), to get marker coordinates in terms of the base coordinate system, but not for sending unlabeled data.
0:12:27 Rigid Body Hash Codes: QTM uses hash codes for rigid bodies when streaming data, which are mapped to rigid body names in an XML script.
0:15:30-0:16:00 Centroid streaming for the rigid body: The presenter can now get the centroid of a rigid body from QTM, for the center position and the rotation of that rigid body.
0:16:55-0:17:20 Frame Rate Issue: Rigid bodies slow down the real-time streaming rate, dropping from 300 HZ to 25 HZ. Disabling rigid bodies allows real time rate to go back up to 300 HZ. The presenter will be asking his contact at Qualisys about this.
0:19:08-0:20:00 QTM Helpfulness: QTM is incredible and that if there is something I want qtm to do it's pretty damn likely that it can do it it's just going to be a bit of figuring out how to do that and to be honest.
0:20:00-0:21:06 Marker Symmetry: Having all the marker points symmetrical is no longer necessary. However, at least one marker needs to be kept on the origin.
0:21:06-0:21:25 No Space Mouse support: QTM does not support space mouse functionality.
0:21:25-0:23:46 Virtual Markers: One can add virtual markers within QTM to assign markers in places where it may not be physically possible.
0:24:00 Redesign of the baseplate: The baseplate has been redesigned, and was previously held on with magnets is now screwed on.
0:25:00 Redesign of the marker holders: The marker holders have been redesigned, which allows more randomization.
0:27:00 Issues with the code: Issues have been identified on the code.
0:30:10-0:31:50 Base Manufacturing Challenges: The presenter faces challenges in manufacturing the base with required precision, considering SLS resin printing and potential warpage.
0:31:50-0:33:30 Twisty Cables obstructing markers: The twisty cable that is wrapping around is actually blocking the markers sometimes, so the markers are having a harder time being read.
0:33:30-0:40:44 Cover for the base plate: A fabric cover was created to kill two birds with one stone and cover the base both during calibration as well as for regular use to save it from dust and cat hair, but Alby the cat likes it as a cat tent.
0:41:04 Finding the Center of the Platform: It is hard to lock onto the platform because it is difficult to say where that point is so we can go into the motion capture software and we can say yep platform 7 or whatever it is, has these coordinates with respect to the platform centroid.
0:41:04-0:47:44 Determining Platform Coordinate System: The presenter aims to accurately locate the platform's centroid in the tracking software, but the imprecise geometry of the platform and its attachments presents a challenge.
0:44:45 Hope for at least horizontal carbon fiber tubes: You should hopefully work which is to kind of like hope that at least these three carbon fiber tubes the horizontal ones are in fact more or less horizontal and if they're not then they kind of average each other out and are equally I don't don't know they like yeah that together if we kind of like average all of those out then hopefully this position here is in fact directly over the centroid of the platform so now I can say okay this point here is however many millimeters above the actual centroid I say yep that is where the centroid is of course.
0:44:45 Move the platform and track it with the motion capture: Here we have a couple plots the main one let's let me just show you let's get rid of these rotations for now and let me move this away okay yeah so this top plot is the Y position against the X position so this is like if we were looking down from above from the roof at what jugle bot was doing then this is the pattern that we would see it follow and down the bottom here we've got the Zed position and we can see as we move through well you can just see straight off that the motion capture which unsurprisingly is the messy the Red Data is kind of far off the commanded data so the commanded one is the blue nice clean crisp lines beautiful that's what we want to see oops sorry but unfortunately.
0:55:30-0:56:04 Significance of the Plane: The plane where the legs attach has particular significance, because that is exactly the plane at which the legs attach and that plane is important for the inverse kinematics of moving the steart platform around but.
1:00:08-1:00:57 Stewart Platform Paper: Even though the biggest error in any of the geometry of the platform was something like half a millimeter like one joint was like half a millimeter out of place or like all the joints were a little bit off but the biggest error was around half a millimeter I think the platform as it was moving around experienced the largest error of almost 15 mm and when I read that I was like oh crap because my robot.
1:01:00-1:03:31 Addressing Geometric Inaccuracies: With all of the steps mentioned thus far, it should help address some inaccuracies.
1:03:31-1:06:41 3D Printing the base: the base needs to be spot on maybe we'll talk about why it's maybe in a bit um but yeah I I really want this to be spot on and so yeah Waring is not good they've suggested that if I increase the wall thickness cuz right now it's it is kind of kind of thin it's like 2 Ms I think yeah 2 Ms I think the around yeah they said if I increase the wall thickness to like five I think they said then it will have less chance of warping but I'm like less chance is not no chance and I kind of want no chance so I don't know.
1:06:41-1:08:00 Need for more sophisticated tools for this project
1:08:00-1:10:30 Discrepancies between test and code:
1:10:30 Different errors causing bad accuracy: To James' point I do think it's fairly likely that not all six legs are the exact same length when they are as short as possible especially because at the bottom of each leg is just like a little foam piece that I'm using to kind of like when when the legs go through the Homing sequence they run down until they hit that foam and then the motors like can't go any further and so the current in the motor will Spike the O Drive detects that spiking current and then says yep I'm at the end stop I don't want that Collision to be like a rough hard rigid Collision so I've added that piece of foam in there but over time those legs have homed slightly differently for whatever reason and some of those foam pieces are more squished than others which of course means that all the legs when they are homed are all slightly different lengths not to mention the fact that again this is all made by hand and so errors can compound.
1:13:55 Goal to have an accurate robot: The plan here is to have the robot this sad face was a bit of a meme just for the sake of the thumbnail I'm kind of hoping that maybe no probably not next week but at some point in the future we will have a a happy face thumbnail with the red and the blue lines perfectly um superimposed on each other which will be great.
1:15:35-1:16:00 Brute force wont work: this would functionally not work I think it would I would have to be leaving my computer running for like six months to finish this simulation um but a more sophisticated version of this I think a more clever version of basically yeah just trying every single combination of those 42 geometries and then seeing which of those combinations matches the actual outcome the best and a key Point here is that we need to know how long all of the legs were in that movement so we say okay the robot has some given unknown geometry it moves the legs to be these lengths that we see here so it's moving the legs to be in these in the at these given lengths and the end result is that the platform moves to wherever we see here so then we say okay if we move if we have this given geometry that we are um that we we are analyzing right now so some version some variant of what it of what it it's supposed to be if we have that combination of geometries and we move those legs to be the given lengths then does that match up with what the motion capture observes if well once we find a geometry that matches really really closely then we can say with some high level of confidence that that is in fact the geometry of the actual robot and then we can say in jugle bot's code we can say Okay base node zero is at these coordinates bass node one is at these coordinates and then so on so on so on for all of them we can say the default lengths of all of the legs are this this and this and then every time we move the robot in the future it will use those coordinates as the geometry of the robot and then we should be getting very precise.
1:15:30 and 1:16:04 and more importantly, there are 42 unknowns: for every single base those 42 positions are all of the base nodes of which there are six have an XY Z coordinate so that makes up 18 all of the platform coordinates of which there are six have an XYZ coordinate which makes up another 18 s of to 36 and then the remaining six are the like shortest length of all of the legs so when all of the legs are as compressed as they can possibly be how long are they now.
1:17:54 What this means is that the robots absolute position is not known: if the robot goes to 100 mm then there is a good chance that it will not go to 100 mm and rather all over the place. This will need to be recalibrated with the new updates to the system.
1:19:51-1:20:15 How to detect if the ball is being carried within the hand, have a ring of the same marker tape within the system, so hopefully we will know. The camera with the tracking should know of this or be close. If not we can say it's not working. If not we do not trust.
1:21:00 Pressure plates?
1:21:50 div it spots good idea! - James Says " "suggesting model divots for the marker stickers into the print so you know where they are that's a good idea"
1:23:17 - A big relief that the system has what he needs: A big relief that the system has what he needs "I was kind of getting a bit like ah like this isn't really doing exactly what I want you know I'm kind of having to like push like force a bit harder to get what I want out of this but it turns out it was just a skill issue and Philip my guy came to save the day."