Відео

What kind of backlash do you get?

Maybe you could crimp some ring style connectors on the end of the sychromesh.

Could this design concept work for an affordable, household motion sim?

Ok I've got to ask for someone for intelligent than myself to ask. When i first started following this thread i really didn't expect this. I love it but maybe I'm too sci-fi at heart. I was honestly, seeing the size of your actuators, thinking you would make two to four specialised "HarrisonLow-Stuart" platforms. I imagined the base plate to be utilised, installed as the start point to legs or arms. Figuring the self leveling aspect was being utilised instead as an ankle to keep the body balanced through a step motion. Honestly, i clicked this wasn't the original goal. But to my question to the intellectuals here. It seems theoretically do able, maybe the Stuart platform would need outer plating like in the images just to keep the joints from flexing to far. Idk😂 But it made me think

Главное вовремя отбежать, когда они падают и начинают хаотично тыкать в разные стороны. :)

*PDJ Weekly Livestream #41 - Motion Capture Update: Refining Ball Tracking for Jugglebot* * *0:45** Calibration Tools:* Calibration tools are now stored near the door, minimizing interference with the motion capture cameras. * *1:52** Qualisys Site Visit:* A Qualisys representative visited to help optimize the motion capture setup. Key takeaways: dirty markers are not a major issue, "active filtering" can mitigate sunlight interference, and camera masking can prevent camera cross-interference. The L-frame for calibration can be positioned freely, not necessarily on the ground. * *10:42** Improved Ball Tracking:* Ball tracking now handles multiple balls and minor occlusions with accurate landing predictions. Playback data shows landing predictions vary slightly even with identical throw repetitions, highlighting the inherent variability in robotic control. * *18:20** Kalman Filter Timing Issues:* A timing discrepancy between the motion capture data stream (300 fps), the robot's processing rate (200 fps), and the Kalman filter's assumed rate was causing fluctuating prediction accuracy. Correcting this significantly improved prediction consistency. * *21:39** Alternative Tracking Methods:* Discussion of using velocities and accelerations for motion classification instead of quadratic regression. Noise amplification from derivatives poses a challenge. * *34:14** Computer Vision for Tracking:* Normal cameras with depth estimation are a possibility but unlikely to match the sub-millimeter accuracy of the motion capture system. Depth camera data was previously found to be too noisy (around 10cm). * *40:04** Call for ROS2 Developers:* Request for assistance with ROS2 performance characterization and analysis to identify potential bottlenecks. * *43:37** Last Stream of the Year:* Holiday greetings and announcement of a break until January. * *46:37** Tracking Data on Github:* Motion capture data (three-ball and five-ball juggling) will be uploaded to Github for community access and experimentation. File formats under consideration: TSV, FBX, and JSON. * *48:19** Alternative Reflective Material:* Testing reflective string as a potential alternative to retroreflective tape for covering the juggling balls, with plans to crochet a sleeve. * *50:28** Ball Characteristics and Tracking:* Speculation about whether the sand-filled juggling balls' unusual flight characteristics contribute to tracking challenges. I used gemini-exp-1121 on rocketrecap dot com to summarize the transcript. Input tokens: 28415 Output tokens: 502

could you order some aluminum spheres anodized or powder coated w/ a tough retroreflective surface?

I think you can reduce the complexity of determining whether a ball is on a parabolic path by ignoring any tracked object below the top possible height of the robot hand. Nothing above that point will be anything other than a thrown ball, and nothing below that point needs tracking because by then it will either be caught or not. This is how is human jugglers track the balls when we look up at the pattern. We don’t pay any attention to the balls in our hands.

I missed the start of the stream and was only able to join the Live for a few minutes, so imagine my surprise when I came back later and was able to watch it properly from the beginning :)

chosing a website that u can check without needing to make an account is such a based thing, it makes all that informations accessible from internet without any walls hiding things away

Position one of the cameras directly above the jugglebot looking straight down to eliminate high-throw dead spots?

16:42 I did come back to watch the video! I used to do some work with four-point tracking for drones and small autonomous machines/devices in an enclosed volume (gymnasium) so getting to watch you calibrate your volume in near-realtime was fantastic! Gonna watch the rest of the video now :)

Congratulations on your engagement!

To detect projectile motion its sounds like you might be doing something more complex than needed. Just use the velocities (vectors) of the ball over time (could be from the your kalman filter), and use their differences between them to compute the acceleration vectors. Average over your time window, then subtract gravity from that and look at the length of the resulting vector. If its large, its not projectile motion.

One note on the hand competition is that I'd probably have the submission period open for around 2 months or so. Do you think you'd be likely to make a submission?

Why dont you use 65lbs fishing line, with dyneema? Same size line as you have, much higher break strength

I understand that your thread is keflar. Dyneema is a lot stronger and lighter. I don't know about bend (around a pulley) fatigue compared to kevlar.

How hard would it be to scale this down to neck or wrist size?

The motion control stuff feels like something from a stuff made here video, but I'm not sure which one. He has had a bunch of hit a target type projects.

As for breaking strength of the line, if it broke at your knot, then it pretty much always does and if it's just a bowline or similar, you lose about 40-60% right there as far as I recall. Then again, still far from the state strength. Splicing it, if possible, will pretty much get you to the regular line's own strength. Meaning, it wouldn't break at the splice. But splicing really thin line is tricky/sometimes near impossible. I guess you have already looked into various fishing knots? They tend to be the strongest around.

I may be wrong but couldn't you lose one pull on the bottom..? In the instance of the blue bottom, stand the single pulley up along the CF tube so it's only job is to turn the string 90d from the motor pulley to the axis of the tube. Basically, keep the one under the CF tube, but turn it so it aims towards the pulley.

*PDJ Weekly Livestream #39: Jugglebot Motion Control Update and More* * *1:06** ROS2 Remains:* Despite limitations, ROS2 will continue to be used for the project for now, allowing for a deeper understanding of its capabilities and potential challenges. * *3:09** Hardware Upgrades Reduce Wobble:* Newly installed carbon fiber nylon parts significantly reduce platform wobble during high-speed movements, as demonstrated through slow-motion footage. * *11:27** High-Speed Camera and Lights:* The livestreamer uses a Freefly Wave camera with an IRIX 150T 3.0 macro lens and Amaran 300C RGB lights for slow-motion capture. * *15:10** New CAN Connector Mount:* The CAN connector mount has been redesigned for stability and features a red color for easy visibility and safety. * *18:09** E-Stop Functionality Restored:* The emergency stop (E-Stop) system is working as intended, enabling safe and efficient recovery from errors during various stages of robot operation. * *22:45** Careful State Transitions:* Transitioning between different robot control modes (e.g., space mouse, shell, juggling) requires careful management to prevent abrupt movements and potential damage. * *24:49** State Machine Improvements:* A new state machine with standby idle and standby active states, along with a generic control mode state, facilitates smoother transitions and future expansion with new control modes. * *37:03** Motion Control Insights:* A paper on quadcopter trajectory generation provides inspiration for a new approach to Jugglebot's motion control, potentially replacing the previous "catch plane" concept with a more flexible, path-based system. * *39:41** Ball Trajectory Simulation:* A new simulation tool helps visualize ball trajectories and their intersection with Jugglebot's workspace, aiding in the development of path planning algorithms. * *46:39** Path Optimization Considerations:* Potential path optimization criteria include minimizing leg velocities/accelerations, minimizing distance traveled, and maintaining platform stability. * *49:44** Open-Loop vs. Closed-Loop Control:* The impact of open-loop control (relying on throw accuracy) and closed-loop control (using motion capture for ball tracking) on path planning will be explored. * *56:02** Upcoming Stream Break:* There will be no livestreams for the next two weeks due to a planned trip. I used gemini-1.5-pro-exp-0827 on rocketrecap dot com to summarize the transcript. Cost (if I didn't use the free tier): $0.04 Input tokens: 28466 Output tokens: 523

It seems to me like having completely computed paths every catch is quite extreme. Especially once you have a network of these things and a decision on one affects all the others. Couldn’t you pre program the whole sequence and then just do slight deviations to manage real life variation? Seems like that is what a person does too, muscle memory plus a little bit of correction.

If you need to get stiffer and stronger, check out this new material. 3D printing nerd just did an interview with the creator. Amazing stuff but expensive at 275 USD for 500gms TULLOMER-500G Replacement for aluminum, steel, magnesium, PEEK, PEKK, PAEK, ULTEM, Nylon CF. Key Features: Ultra lightweight High stiffness and strength Superior dimensional stability No need for annealing

Hey, Cool project, just discovered your channel. Couple of thoughts, the line you use is quite rough which could potentially contribute to the precision issues you're facing. There's a product using a dimensionally controlled kevlar cable, it's a bit thick for your application perhaps (2mm), but a lot more round. For the TPU and bearing sleeves, UHMWPE could also be an option, also thin silicone sleeves with a high shore value could. Last thing regarding the precision, I haven't seen any tensioners for the strings and how do you control the stack up of the string in the spool ?

About the hand, you are mentioning possibly using a cone. How much does the weight of the hand affect the speed of the system ? I guess thats one more answer you would get with a testing rig! Just putting it out there : an alternative to a cone that would be lighter would be where the side of the cone is open ? I am not sure what is the name of the shape, but check pictures of sputnik, remove the ball at the bottom, put your current hand instead and maybe add more rods ?

RE: ROS, I'm with you on that. My humble opinion is that you should probably not think about it until you REALLLLYYY have performance issues. In which case you should measure first and then optimise IF needed. For sure there would be a cost in switching to a different option, should you do it later as opposed to now, but is ROS really the biggest problem today ?

im sorry i dont have time for the whole video; does your linear actuator have capability to "relax" when unpowered and shrink when to much opposing force is on it?

Have you looked into road marking compound? It survives cars and trucks driving over it, so it´s durable. The white compound might be translucent in thin layers as well.

Hi Harrison, the async benchmark is more like 49 times faster than ROS Python not 200 times 😜 In any case, I don't think performance is as much a ROS issue as a language issue. ROS C++ would give you much better performance overall, not just in messaging. YMMV depending on what else your code is doing but a 10 to 100 X increase is possible. Obviously your appetite for coding in C++ would be a big factor. Python is much better for rapid prototyping. I don't know what is involved but maybe porting from ROS Py to ROS C++ if and when performance becomes an issue could be considered.

Maybe you could use a tube to throw the ball in the hand? You could make different holders for the tube and simply throw the ball in on one side of the tube. This would be repeatable and also the velocity of the ball would always be the same.

Sad I missed the Live, but definitely wanted to chime in with regards to ditching ROS. I was working on my first-ever inverse kinematics project that involved computer vision, trying to do it all with libraries, cores, and all kinds of hodgepodged things talking to each other, brute-forcing as much of my own code to get it to work as possible. Unfortunately, it was a chaotic mess, and some people pushed me toward ROS. This was about 15 years ago, when things were very different, and when I was very different, so I can't speak for modern iterations and such, but: it was helpful in an organizational sense but limiting in a functionality sense. I could use ROS to achieve most of what I needed, with generous help and assistance from ROS's community, but when ROS began to hold me back, I was also held back by the community. After years of struggle and making the decision to finally ditch it, I can definitely say, ROS is a great way for beginners or people who are relatively new, to get an understanding of structure and good technique, but beyond that, it feels like trying to build a microscopic precision sculpture out of Duplo blocks. Yes, it's easier to have your hand held, but it just doesn't work out in the long run.

Use a modified cable wedge clamp design with a set screw to lock cables.

Very cool! Really fun to see you work the logic of the state machine... Do you mindmap your thinking in an app? I use obsidan canvas as it is so easy.... aka the issue of refactoring the drive data to its own topic? Thanks and keep at it!!!!

Does the ODrive have any latching user-settable unassigned flags? The teensy could write to the flag when homed.

Another thought on making the balls retroreflective (discussed around 1:14:00) that I should've mentioned is to spray the balls with some sort of glue/resin and coat them in tiny glass spheres (since these spheres are what usually makes retroreflectivity work). I'm interested in this idea, though I have a few concerns: 1 - This will be messy (and possible hazardous if the balls are small enough) 2 - The balls will likely wear off over time, affecting both tracking ability and the messiness noted above. 3 - From what I've read, the balls need to be 'submerged' in the glue/resin just the right amount for retroreflectivity to work. IIRC around 45-55% submerged is the ideal amount, and I'm not sure if they're even retroreflective at all if they're fully encased. I'd definitely want to do some testing on this before coating an entire ball, though I'm unsure if it's even worth going down that route at all... 🤔

Very impressive project! 👍

I'm not into programming/coding, but, to me, this episode was way less interesting than the preceding ones.

I have no experience with it but retro reflective spray seems better to me since you don`t change the the tactile feel like you would with tape. Or because I expect the wraping of square tape around a sphere something realy difficult and not worth investing time in. You propably spend more time thinking about it so I'll just shut up. Greetings From the Central European Time zone :) PS: or any kind of paint.

yo, imma need those design files. my girl saw some of the clips of the actuator moving and had some ideas, so now i need to build this... its an emergency... =)

I'm probably missing something as this is the first video I have watched on this project. But the three pulley system you use from the motor to get the right orientation. You are using it to translate the liner motion 90 degrees to the motor. Why can't you just mount the motor 90 degrees instead. I mean it looks slick all inline.

You could terminate the cables using eyelets. That would also make them very easy to swap out.

Logs slow things down, sometimes that fixes bugs when things are being done too fast / close together.

brake resistors are really job specific. Got to do the calculation for the brake power. This is the weight of the load, time to stop and a few other things. I have always went with what the manufacturer recommends. Industrial Allen Bradley (Rockwell) will have a chart/calculator to tell you what size drive and brake you need. This also keeps the parameter settings straight forward by using all the same system. Also, if you start getting into higher end drive systems some of them are modular. Allowing them to be all hooked together and then controlled through tcp or other standard protocol.

To elaborate on what I talked about at 9:33 and 13:30 regarding MonitorStates missing messages due to being overloaded(?), I'm still very unsure of why this happens, and would appreciate any thoughts on why this might be happening! 🤔

Can you disconnect the +5V wire on the USB to the teensy, so that it will only get power from the 12V supply?

One general best practice in software is to make invalid states unrepresentable. For example in your case you have a few flags you set to indicate progressing through the various phases of initialization. If you never want to have to think about the case where the later ones got set without the earlier ones, you can replace the all the flags with a single enum with values something like uninitialized, homed, and leveled). This, similar to your state machine's state, ensures the system is reporting exactly one coherent state (what ever the enum says), and can't even express invalid combinations of them (like leveled but not homed)

Go online and find a game dev and Ai programmer. Make a physics game to simulate the juggling bot and the controls with the same camera input. And let it learn how in parallel processes

Use nylon for the bearing sleeves, and make them concave to suit tube. Clamp the chain.

very cool, but one thing I didn't think was cool is that the actuator has a slightly large area in terms of its total diameter, and it seems like it would be complicated to adapt it for other activities... how would I fix the actuator for different activities... anyway, it's very interesting, mainly because it uses all carbon fiber, which I think is a sensational material.