Not exiting? Pfft. That was interesting as hell watching your development process, wild that those motors cause that much noise! Can bus integration will be very interesting.
It's like Minecraft or Dwarf Fortress. *Losing is fun*: every failure is a learning experience and every end is a start of something new. Seeing different attempts and failures is maybe the most interesting part of R&D.
the reason it's so noisy is that you have wrapped the signal cable *around* the motor cables! the shielding protects mostly against EMF in the radio and above frequencies, not against magnetic flux in the low frequencies of motor drivers. this effect is even more severe with motors because their high inductive impedance means that any in-line capacitance allows large, periodic, asymmetric current flows. the setup is not ideal to start with but, these wraps almost guarantee an unusable signal. cool project by the way!
The noise is probably induced in the loadcell itself due to the motor noise , the digital pwm signal in a shielded cable won't be effected that easaly.
Two steps forward, one step back! In case it all gets to be too much of a chore, I have sent you a "quick-fix" package... One house-trained black labrador, and a large can of yellow paint. :o) Always look forward to your videos, thank you so much.
this man is a legit genius. i cant believe one man can do so much. how many times have poor old boston dynamics been rejected when they come knocking at his door with a job offer i wonder?
I like this idea a lot more then just having an on/off signal from a foot sensor, to bad the motor cross talk is a stronger signal then the one your trying to measure. I'm sure you'll figure this one out.
I feel the load cell is going to be helpful for having more info on how the weight is being distributed among the legs. Another thing I just noted is that the legs don't spread outwards individually (as Boston Dynamics's notably do). Having the legs spread helps making the center of gravity always be centered in the triangle it forms with the feet. It also allows crossing the legs when the dog is moving/falling sideways to try keep the center leveled/high up.
You can linearize the non-linear sensors if you can do a bench calibration. You could collect all the force sensors into a teensy talking to the main controller. Chris
Hey James! For some reason, this kind of driver that comes with the load cells are set to work at a low frequency, but there is a pin in the IC that can be shorted to vcc and that alows the sensor to wortk much faster! Its quite tricky to do because you have to lift the IC terminal from its pad, but the results are realy worth it!
Vid was really exciting, give yourself credit lol. When you're investigating CANBus can you do it in a folow along type style please, plenty of us have never used CANbus and I'd love to learn along
I'm looking forward to the CAN bus episode - especially details on the capacity of the bus - update rate, number of devices, etc for robotic applications
Just an idea.... When doing the trot gate try moving the foot to where the floor should be **at that angle** so it gets the foot on the floor to stop the tip but does not try to correct the tip until all feet are on the floor.Then return to upright with all four feet on the floor. That should stop it punching the floor and knocking it over the other way.
@@TheLordFisch Took them long enough, FRC (First Robotics Competition) which is a international robotics competition for high schooler's has been using can bus religiously for nearly half a decade, PWM and other standards like I2C, RS232, etc., are still used but Can Bus is a beast.
Check to make sure the cable shield is connected to the motor supply return. If the Trinamic board generates it see if you can find the common. Should eliminate the crosstalk. Also make sure the shield is only tied to that return on one side
The power noise could be from a shared power source and coming from the Arduinos power lines from within the shielded cable. Maybe try a UBEC to power the Arduino. Awesome video. Keep it up!
Cool video James! I'm in the middle of a 20 loadcell project myself and I'm using the hx711 like you were originally. It can go up to 80Hz if you jump pin 15 I think. Would be interested in how you implement the code in case you read multiple loadcells (4+) by the same arduino. I've found that pretty much all libraries for the hx711 can only receive information in series and the only library multihx711 that supports receiving data in parallel from multiple load cells doesn't have useful functions like tare etc.
You're sending the signal as a voltage and are perhaps reading it on a high impedance input port. You might consider putting a load resistor in parallel with the input port. This will increase the current of the signal so that it doesn't drown anymore in the motor induced current. Choose the resistor such that the current is as high as possible, without overloading the transmitter.
Wow, really interesting ideas there James. I'm not sure how spongy the paw is, but any sponginess will skew the readings from the pressure sensor I expect. I don't know if that matters, but thought to point that out. I'm wondering if it's worthwhile to do much more processing of the pressure data in that teensy - distributed intelligence style, muscle memory etc. I think your suggestion for canbus is the perfect solution, but its data rate is quite low relatively speaking when there are other devices on the bus, so local leg intelligence will be required anyway.
The channel GreatScott! here on TH-cam has a great intro video about CAN bus basics that I would recommend to anybody interested in starting to learn about the topic. Also, was that shielded cable also twisted pair? If it wasn't, that would explain the motors inducing current even though it was shielded. It is also possible that you needed to ground only one end of the cable's shielding, typically at the sensor end, depending of the specifications of the cable. If you do go to CAN bus, it uses a twisted pair of wires for its signal wires and is also highly resistant to interference from induced currents and RF noise because of the protocol used and the specs for the wiring for it. I would still suggest more biomimicry on this foot. At least two toes that also feel when they touch the ground and ate mechanically attached to the primary pressure sensor you have already added to feel for ground contact. Each of these three points, the two toes and the ball of the foot, need to be able to determine pressure roughly and absence/presence of ground. What you have now should be able to walk on flat, even surface ground, but trying to deal with anything more complicated that that will prove impossible, even with a gait. Absence of feeling or loss of toes on feet beyond feeling for three points and overall pressure (this is the load cell on the leg for OpenDog, but is the ankle in animals) makes walking on anything beyond a flat surface incredibly difficult even for animals and humans. If the coordination task is difficult/nearly impossible because of random variables for an organic creature to walk on, it is going to prove to be a very difficult to impossible problem a robot. There might be a statistical prediction method (simple AI) using just the leg load cell and the basic switch for ground presence , but this would be moving away from using the programming model of movement that uses input about the presence and shape of the ground to adjust the movement model with simulated springs in the legs to a back box of AI that may work but also might tear the robot apart since it feels no pain/senses no damage to stop it from twisting itself into a pretzel. To be honest, I have never built a robot and have only programmed a few. I do understand the programming and mechanics of how OpenDOng works because of college level classes in electronics, automation, and mechanics. I also have taken programming classes in Assembler and for PLCs. This includes understanding how to set up a PID control. I also have a rough understanding of bio-mechanics from raising and working with animals as I grew up. Noticing a change in gait on an animal could tell you a lot of what might be wrong with it and if/how it could be fixed. This background along with some reading on the topic of how animals/humans walk is where my advice is coming from (I looked into this topic when I had some walking problems to try and figure them out with my doctor. It ended up being a medication side effect, but I learned the basics of the topic).
have you made sure that the interference isn't coming from the loadcell amp? Since it's job is to read nanovolt changes in in the signal from the strain gauge, might be susceptible to the motor interference.
Estimado la solucion para este problema es 1)gndA y GNDD solo deben ser independientes , masas de alta calidad, se unen en un punto 2) otra alternativa es que leas la celda de carga como lo estas haciendo pero colocar un microcontrolador pequeño con una salida 485 y generes un bus de comunicaciones donde todas las patas se unen a ese bus y puedes así leerlas en la cpu principal , estos mismos micros podrían levantar todos los sensores y activar los actuadores "inteligencia distribuida" donde exista un procesador principal y varios secundarios con esto minimizaras el cableado y las interferencia que producen los diferentes controladores , espero haberte ayudado en algo y muy buen canal tienes . Saludo desde Argentina, Marcelo
Very nice video. I'd put all the low signal amplitude stuff away from the motors and odrives inside shielded can close to the sensor. It should help to reduce noise.
The noise goes away if I read the data from the Teensy on the leg with the motors powered up, so it seems to be on the line from one Teensy to the other.
i dont think that type of load cell is designed to work like that. the wheatstone bridge is setup so that strain gauges on the same side of the load cell undergo "opposite" deformation: 1 in tension the other in compression. as you have it both strain gauges on the same side are in either tension or compression. so i think you are actually cancelling out a lot of the signal and essentially amplifying noise. for measuring simple bending and compression you only actually need one strain gauge directly on the rigid Al, but using the resistance bridges offer advantages; look at the "getting started with load cells" spark fun page and the therein linked "all about circuits" page. it may be possible to rearrange the wiring on the wheatstone bridge on your load cell so that it works in your setup.
I agree that the load cell setup is a bit wonky in this prototype. I think it would work better if you directly epoxy the strain gauges onto the top and bottom of the extruded aluminum.
Getting closer :) Any *differential pair based* communications bus should get rid of the induced noise on the cable. I2C is absolutely horrible for long cable runs in noisy environments. In one of my old work projects we had to abandon I2C in a vehicular install as it was too sensitive to noise. CanBus would be my recommendation, but you could try RS-485 / EIA-485 (straight serial or ModBus) as a short term solution. Cheers,
@@jamesbruton RS485 supports multi-drop mode, all the nodes in the network share the same tx pair and rx pair, or even only a single pair if you chose to.
Because you wouldn't be able to tell the weight distribution when the dog is static. Say for example the dog is leaning forward and not moving or very slowly, you still want to know there is more weight on the front paws.
@@thygate If it put the leg down slowly it would be nigh on impossible to tell that it had hit the ground. Why would accelerometers be better than a switch anyway?
@@clonkex No moving parts to gunk up and fail. And I assure you, you WILL see a spike when hitting a stationary object, no matter how low the speed. Find an IMU monitor App on your phone and try putting it down without getting a spike.
@@thygate Tbh I've never once seen a microswitch fail. You'd have to be pretty hard on it to kill it. And sure you'll see a spike when you hit the ground, but reliably detecting that spike could be difficult. A microswitch is just a way more viable and useful solution in this case. Cheaper, simpler, more reliable.
Quite silly that the IC doesn't let you set an i2c address, that's kind of the whole point of i2c, that you can address things without extra wiring, IMO. Looking forward to seeing you explore CAN; never looked into that myself either. :) I do appreciate that you're showing off things that didn't work out, as knowing about stuff that doesn't work out is rather educational as well.
@@jamesbruton I imagine so, perhaps unless you can get away with particularly low pull-up resistances, I imagine that interference would affect I2c data lines too.
If you want to stick with analog data, consider a current loop. That's got very high noise immunity. If you switch to a digital data stream, consider RS-422 or RS-485. Their differential signalling will also have excellent noise rejection. RS-485 *might* be a simpler network alternative to CAN bus (getting a little out of my element there).
You could try differential signalling... in any case what ever bus you're going to choose it has to be differential and properly shielded and the grounding scheme you'll use has to be properly made (star grounding). You need to be sure you're properly filtering your power rails as well with ferrite beads to cut down any noise that creeps through there from the motors. Any cables you run to your motors have to be shielded as well etc. Designing these things isn't easy or simple :D
I would repeat the test with aluminum foil. Wrap it around the cables and in another test wrap it around the sensors by the lower half of the leg to see if they need shielding from the motors. Maybe you just need more shielding. Also, I don't know if you have any ferrite noise choke clamps laying around but maybe you can put one around your data cable to help.
Noisy AC is a pain to deal with. There are some effective filtering approaches used in the RF arts that may help you. Remember AC and DC can be differentially blocked by capacitors, transformers, and filtered using simple arrangements of chokes and cheap passive components. Might be helpful to investigate since you are radiating RF due to the power levels and high frequency drives. You have an RF problem :) CANbus would surely help, but it might still be problematic for data feeds on devices that can't use it.
If you hide the motors in a shielded box, and use a gear system to transfer the power to the legs instead, you'd fix a lot of these signal problems... Though perhaps that would be overly complicated. Maybe just wrap some extra foil around the bits causing interference?
The issues goes away if I get the data from the Teensy on the foot, even with the motors powered up, so it's definitely getting introduced on the line between the two Teensys.
I think the noise is affecting your load cell amplifier on the analog side, not the digital communication side. Try shielding it by wrapping with aluminium or copper foil?
I like your thinking, but I know fibre optic cables can be fragile. I'm not much more experienced, however, so there might be some hyper flexible, tough cable out there.
I wonder if it would be a good idea to send signals on wires made of stainless steel or carbon fiber or something that has higher resistance than copper. Might reduce induction effects. And since there's basically no power needed, it shouldn't result in enough heating/energy waste to matter.
I wonder about a shock absorber , you could measure the linear delta in the compliance . keep up the good work. looks awesome. id love to do smaller version I could follow along and make myself
I work professionally with load cells and strain gages. This is not how to use those load cells. It'll cause sensitivity to things you don't want. I'm glad you came to ditching the HX711, it's literal garbage. The way that load cell is setup is kind of like a parallelogram but with contraflexure (Thin element bends at both ends, one causing compression and one causing tension). That cut out has 4 gages in a wheatstone bridge on one side where one half of either element (top or bottom) is in compression and one half in tension due to how it flex's with the cutout. IF you mount it like that it's acting like a poor differential bending beam and you're constraining the contraflexure element. This becomes highly unpredictable and force calibration could change with bolt tension. So remove and remounting could cause huge changes in calibration. It will also decrease sensitivity and cause new sensitivities to things you likely aren't aware and potentially making it unstable. Really, you might need a custom design because most load cells are 1 direction, normally perpendicular to the beam. However depending on what angle the leg is at, this might tell you only a partial story. A dual axis design would be better but few people deal with multi-axis load cells. I do, but I'm a rare case.
Hi James Instead of using CANbus have you considered ROS (robot operating system) ? This would require a Raspberry Pi but is intended for this type of application.
CAN bus is a standard for data transmission (like serial), whereas ROS is for control of a system as a whole. I agree that using ROS would be a good step forward, however, and to top it off it can work with arduino so long as there is a Linux computer to send the appropriate messages.
Great project. You can use multiplex to deal with limited i2c addreas- might scale nicely to 4 legs. Pca9617 is handy to boost i2c signal. With your pwm signal did you put a terminating rc network in? Also a cheap mcu can act as an i2c slave.
If You need any help about CAN bus let me know ... differential signal will cancel your noise for sure . But maybe you can manage without it just with some smart wiring. Anyway there are plenty available libraries, you just need to choose some transceivers. Dunno if your uC have a CAN controller already implemented though ... if not you should use some other differential protocol.
mabye you could use fibre optic, I have a tesla coil and need to use fibre optic cables for the interupter because of all the interference the coil makes
Are you sure you are using the gait? There are six style gaits, and I think you are trying to use one that has forward motion already to help the balance. A tri-leg stance is not feet straight across but two wide apart and one in middle.
CAN bus! Check out CANopen used in robotics. I found it difficult to find any standards for it though. it looks like all messages are home grown. or you have to purchase message stacks. J1939 is a standard in wheeled vehicles but not very good for your application. Look forward to seeing what you come up with as far as the CAN buss!
What if you send the signal with a twisted pair of wires, one sending the signal, and the other sending the reference "ground" voltage, then you subtract the "ground" voltage from the signal to filter out the interference?
This is not always true. Grounding one end should suffice. If you ground both ends of the shield but the ground at each have a different potential it can cause all sorts of problems.
The noise may not be in your shielded cable but from the load cell itself? Considering the level of amplification involved I would look at that end first.
Not exiting? Pfft. That was interesting as hell watching your development process, wild that those motors cause that much noise! Can bus integration will be very interesting.
My thoughts exactly, Identifying issues is part of the process not to mention the other 15 minutes of engineering.
I mean, those are motors meant for speed and not necessarily precision. Approach to using them in a high precision application is really interesting!
It's like Minecraft or Dwarf Fortress. *Losing is fun*: every failure is a learning experience and every end is a start of something new. Seeing different attempts and failures is maybe the most interesting part of R&D.
the reason it's so noisy is that you have wrapped the signal cable *around* the motor cables! the shielding protects mostly against EMF in the radio and above frequencies, not against magnetic flux in the low frequencies of motor drivers. this effect is even more severe with motors because their high inductive impedance means that any in-line capacitance allows large, periodic, asymmetric current flows. the setup is not ideal to start with but, these wraps almost guarantee an unusable signal.
cool project by the way!
The noise is probably induced in the loadcell itself due to the motor noise , the digital pwm signal in a shielded cable won't be effected that easaly.
@@pimbakkerpim this.
This is freaking awesome. The amount of work that goes into one video, incredible.
Don’t apologise, you covered a hell of a lot in this video 👍👏
12:04 - that length of all-thread seems to be bending. Not sure if that affects things at all but thought I'd mention it in case. Cheers
He did comment on that. He know.
Two steps forward, one step back! In case it all gets to be too much of a chore, I have sent you a "quick-fix" package... One house-trained black labrador, and a large can of yellow paint. :o)
Always look forward to your videos, thank you so much.
this man is a legit genius. i cant believe one man can do so much. how many times have poor old boston dynamics been rejected when they come knocking at his door with a job offer i wonder?
I like this idea a lot more then just having an on/off signal from a foot sensor, to bad the motor cross talk is a stronger signal then the one your trying to measure. I'm sure you'll figure this one out.
Your knowledge on robotics, programming and engineering is insane. I love it
I feel the load cell is going to be helpful for having more info on how the weight is being distributed among the legs. Another thing I just noted is that the legs don't spread outwards individually (as Boston Dynamics's notably do). Having the legs spread helps making the center of gravity always be centered in the triangle it forms with the feet. It also allows crossing the legs when the dog is moving/falling sideways to try keep the center leveled/high up.
Please more videos of Opendog, I think most of your subscribers really appreciate this project!
You can linearize the non-linear sensors if you can do a bench calibration.
You could collect all the force sensors into a teensy talking to the main controller.
Chris
That's quite the friggin build there, mate! Respect!
thanks!
Thanks James for promoting #TeamTrees !!
Hey James! For some reason, this kind of driver that comes with the load cells are set to work at a low frequency, but there is a pin in the IC that can be shorted to vcc and that alows the sensor to wortk much faster! Its quite tricky to do because you have to lift the IC terminal from its pad, but the results are realy worth it!
Vid was really exciting, give yourself credit lol.
When you're investigating CANBus can you do it in a folow along type style please, plenty of us have never used CANbus and I'd love to learn along
The End Of My Rainbow
: I'd like to see that too. The little I have seen of CANBus looks interesting.
yes for CANbus
One of the most exciting videos due to really good explaining 👌
You can epoxy the strain gages on the alu extrusion directly, save yourself the load cells, reduce nonlinearity and offset error.
Also makes it easier to do EMI shielding
I'm looking forward to the CAN bus episode - especially details on the capacity of the bus - update rate, number of devices, etc for robotic applications
Great Scott's channel just did a nice explanation about CANBus.
ive never even heard of a load cell until today this video was my favorite open dog video so far
I just looked into CAN bus myself while doing research for a dual-motor electric longboard project; turns out CAN bus is pretty freakin cool!
Boston dynamics would take this guy in a heart beat
Just an idea.... When doing the trot gate try moving the foot to where the floor should be **at that angle** so it gets the foot on the floor to stop the tip but does not try to correct the tip until all feet are on the floor.Then return to upright with all four feet on the floor. That should stop it punching the floor and knocking it over the other way.
I'm really glad that the biological systems don't have the problem of possibility of tearing themselves apart
Try doing some serious sports without warming up. Ruptures of muscle fibers are not that uncommon.
@@Christian-ev1zu i know the feeling
What we have that robots don't is a zillion little nerve endings to detect problems.
"WHo'se the rigid boy?! :o Who's rigid!? Yes you are! YES U AREEE!!! :D"
Stiff is the word... Erect even...
Happy Holidays !
Yay back to the dog!! We have the faith :)
Love this project. Please continue the series with opendog
Everyone is talking about canbus! I think this is the third channel I've heard talk about it.
bigclive, GreatScott! and now James. We'll see who's next. Probably electroBoom or eevBlog
@@TheLordFisch Took them long enough, FRC (First Robotics Competition) which is a international robotics competition for high schooler's has been using can bus religiously for nearly half a decade, PWM and other standards like I2C, RS232, etc., are still used but Can Bus is a beast.
@@TheLordFisch : Add Skyentific to your list! :o)
I think MIT Mini Cheetah is what started the current CAN bus craze.
dekutree64 CAN in robotics is not new, we have been using CAN for years
You could use 2 light sensors 2 leds and 2 optical audio cables for an optical interface
Check to make sure the cable shield is connected to the motor supply return. If the Trinamic board generates it see if you can find the common. Should eliminate the crosstalk. Also make sure the shield is only tied to that return on one side
The power noise could be from a shared power source and coming from the Arduinos power lines from within the shielded cable. Maybe try a UBEC to power the Arduino.
Awesome video. Keep it up!
Its a shame u stopped working on this now I was looking forward to seeing it walking :(
Interesting as Hell.. I agree... thanks James
Cool video James! I'm in the middle of a 20 loadcell project myself and I'm using the hx711 like you were originally. It can go up to 80Hz if you jump pin 15 I think. Would be interested in how you implement the code in case you read multiple loadcells (4+) by the same arduino. I've found that pretty much all libraries for the hx711 can only receive information in series and the only library multihx711 that supports receiving data in parallel from multiple load cells doesn't have useful functions like tare etc.
You're sending the signal as a voltage and are perhaps reading it on a high impedance input port. You might consider putting a load resistor in parallel with the input port. This will increase the current of the signal so that it doesn't drown anymore in the motor induced current. Choose the resistor such that the current is as high as possible, without overloading the transmitter.
beautiful project
Awesome bro. Great work
hello, the french youtube channel "les frères poulains" recommended your channel. It looks amazing and i can't wait to watch all of your vidéos.
Thanks for coming over!
Always inspiring. Will be working on a full humanoid robotic arm
One of the most interesting parts so far. Tnx James :-)
Wow, really interesting ideas there James. I'm not sure how spongy the paw is, but any sponginess will skew the readings from the pressure sensor I expect. I don't know if that matters, but thought to point that out. I'm wondering if it's worthwhile to do much more processing of the pressure data in that teensy - distributed intelligence style, muscle memory etc. I think your suggestion for canbus is the perfect solution, but its data rate is quite low relatively speaking when there are other devices on the bus, so local leg intelligence will be required anyway.
The channel GreatScott! here on TH-cam has a great intro video about CAN bus basics that I would recommend to anybody interested in starting to learn about the topic. Also, was that shielded cable also twisted pair? If it wasn't, that would explain the motors inducing current even though it was shielded. It is also possible that you needed to ground only one end of the cable's shielding, typically at the sensor end, depending of the specifications of the cable. If you do go to CAN bus, it uses a twisted pair of wires for its signal wires and is also highly resistant to interference from induced currents and RF noise because of the protocol used and the specs for the wiring for it.
I would still suggest more biomimicry on this foot. At least two toes that also feel when they touch the ground and ate mechanically attached to the primary pressure sensor you have already added to feel for ground contact. Each of these three points, the two toes and the ball of the foot, need to be able to determine pressure roughly and absence/presence of ground. What you have now should be able to walk on flat, even surface ground, but trying to deal with anything more complicated that that will prove impossible, even with a gait.
Absence of feeling or loss of toes on feet beyond feeling for three points and overall pressure (this is the load cell on the leg for OpenDog, but is the ankle in animals) makes walking on anything beyond a flat surface incredibly difficult even for animals and humans. If the coordination task is difficult/nearly impossible because of random variables for an organic creature to walk on, it is going to prove to be a very difficult to impossible problem a robot. There might be a statistical prediction method (simple AI) using just the leg load cell and the basic switch for ground presence , but this would be moving away from using the programming model of movement that uses input about the presence and shape of the ground to adjust the movement model with simulated springs in the legs to a back box of AI that may work but also might tear the robot apart since it feels no pain/senses no damage to stop it from twisting itself into a pretzel.
To be honest, I have never built a robot and have only programmed a few. I do understand the programming and mechanics of how OpenDOng works because of college level classes in electronics, automation, and mechanics. I also have taken programming classes in Assembler and for PLCs. This includes understanding how to set up a PID control. I also have a rough understanding of bio-mechanics from raising and working with animals as I grew up. Noticing a change in gait on an animal could tell you a lot of what might be wrong with it and if/how it could be fixed. This background along with some reading on the topic of how animals/humans walk is where my advice is coming from (I looked into this topic when I had some walking problems to try and figure them out with my doctor. It ended up being a medication side effect, but I learned the basics of the topic).
have you made sure that the interference isn't coming from the loadcell amp? Since it's job is to read nanovolt changes in in the signal from the strain gauge, might be susceptible to the motor interference.
Yes it goes away if I read the data from the Teensy on the leg with the motors powered up
@@jamesbruton nice👍 I was pretty sure you would have tested that aswell. Once you get the canbus up and working I think it will be a super solution.
@@jamesbruton If you ground your shield on one side of the cable only, it should help with the EMF noise. Keep up the great work!
Estimado la solucion para este problema es 1)gndA y GNDD solo deben ser independientes , masas de alta calidad, se unen en un punto 2) otra alternativa es que leas la celda de carga como lo estas haciendo pero colocar un microcontrolador pequeño con una salida 485 y generes un bus de comunicaciones donde todas las patas se unen a ese bus y puedes así leerlas en la cpu principal , estos mismos micros podrían levantar todos los sensores y activar los actuadores "inteligencia distribuida" donde exista un procesador principal y varios secundarios con esto minimizaras el cableado y las interferencia que producen los diferentes controladores , espero haberte ayudado en algo y muy buen canal tienes . Saludo desde Argentina, Marcelo
finaly! open dog is back! :)
Very nice video. I'd put all the low signal amplitude stuff away from the motors and odrives inside shielded can close to the sensor. It should help to reduce noise.
Those are some crazy noisy motors! I like the design, though - it's cool and looks like it's also very sensitive.
James I think you will find it’s the vibration of the motors as much as it is eclectical noise.
The noise goes away if I read the data from the Teensy on the leg with the motors powered up, so it seems to be on the line from one Teensy to the other.
i dont think that type of load cell is designed to work like that. the wheatstone bridge is setup so that strain gauges on the same side of the load cell undergo "opposite" deformation: 1 in tension the other in compression. as you have it both strain gauges on the same side are in either tension or compression. so i think you are actually cancelling out a lot of the signal and essentially amplifying noise. for measuring simple bending and compression you only actually need one strain gauge directly on the rigid Al, but using the resistance bridges offer advantages; look at the "getting started with load cells" spark fun page and the therein linked "all about circuits" page. it may be possible to rearrange the wiring on the wheatstone bridge on your load cell so that it works in your setup.
I agree that the load cell setup is a bit wonky in this prototype. I think it would work better if you directly epoxy the strain gauges onto the top and bottom of the extruded aluminum.
Getting closer :)
Any *differential pair based* communications bus should get rid of the induced noise on the cable. I2C is absolutely horrible for long cable runs in noisy environments. In one of my old work projects we had to abandon I2C in a vehicular install as it was too sensitive to noise.
CanBus would be my recommendation, but you could try RS-485 / EIA-485 (straight serial or ModBus) as a short term solution.
Cheers,
yes that's what CANbus is. I don't have any serial ports left for Rs485
@@jamesbruton RS485 supports multi-drop mode, all the nodes in the network share the same tx pair and rx pair, or even only a single pair if you chose to.
how about accelerometers in the feet and watch for the deceleration spike when they touch the ground.. ?
Because you wouldn't be able to tell the weight distribution when the dog is static. Say for example the dog is leaning forward and not moving or very slowly, you still want to know there is more weight on the front paws.
@@didz65 That's why he has the load cell, i'm talking about replacing the switches.
@@thygate If it put the leg down slowly it would be nigh on impossible to tell that it had hit the ground. Why would accelerometers be better than a switch anyway?
@@clonkex No moving parts to gunk up and fail. And I assure you, you WILL see a spike when hitting a stationary object, no matter how low the speed. Find an IMU monitor App on your phone and try putting it down without getting a spike.
@@thygate Tbh I've never once seen a microswitch fail. You'd have to be pretty hard on it to kill it. And sure you'll see a spike when you hit the ground, but reliably detecting that spike could be difficult. A microswitch is just a way more viable and useful solution in this case. Cheaper, simpler, more reliable.
Quite silly that the IC doesn't let you set an i2c address, that's kind of the whole point of i2c, that you can address things without extra wiring, IMO. Looking forward to seeing you explore CAN; never looked into that myself either. :) I do appreciate that you're showing off things that didn't work out, as knowing about stuff that doesn't work out is rather educational as well.
It is, I could use an I2C multiplexer though, but I'd still have issues with the long wires...
@@jamesbruton I imagine so, perhaps unless you can get away with particularly low pull-up resistances, I imagine that interference would affect I2c data lines too.
If you want to stick with analog data, consider a current loop. That's got very high noise immunity. If you switch to a digital data stream, consider RS-422 or RS-485. Their differential signalling will also have excellent noise rejection. RS-485 *might* be a simpler network alternative to CAN bus (getting a little out of my element there).
I would use RS485 but I'm out of serial ports on the Teensy 3.6 due to the six ODrives
wow, thast`s amazing and one thing when robot voices will make phenomenal
GreatScott just did a video about CANbus
You could try differential signalling... in any case what ever bus you're going to choose it has to be differential and properly shielded and the grounding scheme you'll use has to be properly made (star grounding). You need to be sure you're properly filtering your power rails as well with ferrite beads to cut down any noise that creeps through there from the motors. Any cables you run to your motors have to be shielded as well etc.
Designing these things isn't easy or simple :D
I would repeat the test with aluminum foil. Wrap it around the cables and in another test wrap it around the sensors by the lower half of the leg to see if they need shielding from the motors. Maybe you just need more shielding. Also, I don't know if you have any ferrite noise choke clamps laying around but maybe you can put one around your data cable to help.
That's a lot of cable to wrap
Hey. If you get time, do a dedicated video for can bus when you are done with your study on it.
having inside the 3d part the spring so new leg part could totally have flatter spring sensor inside so it could be more accurate
for programmable air could be help stopping and so on based on data from sensor then need to look this repo github.com/programmable-air
Noisy AC is a pain to deal with. There are some effective filtering approaches used in the RF arts that may help you. Remember AC and DC can be differentially blocked by capacitors, transformers, and filtered using simple arrangements of chokes and cheap passive components. Might be helpful to investigate since you are radiating RF due to the power levels and high frequency drives. You have an RF problem :) CANbus would surely help, but it might still be problematic for data feeds on devices that can't use it.
If you hide the motors in a shielded box, and use a gear system to transfer the power to the legs instead, you'd fix a lot of these signal problems... Though perhaps that would be overly complicated. Maybe just wrap some extra foil around the bits causing interference?
Great video mate, have some faith in yourself
nice james!
So awesome to see your development steps here, I learned a lot! I wonder if wire noise filter rings would help as a simple fix for the noise?
I suggest taking a look at you power supply with a scope,, there way be a lot of noise on it. You could try running the sensor from an isolated psu.
The issues goes away if I get the data from the Teensy on the foot, even with the motors powered up, so it's definitely getting introduced on the line between the two Teensys.
Try using the fiber optic serial shield from mikroelektronika
I think the noise is affecting your load cell amplifier on the analog side, not the digital communication side. Try shielding it by wrapping with aluminium or copper foil?
Try fiber optic communication to modules to avoid cross-noise in the system? Just a novice idea.
It's great to follow your creations.
I like your thinking, but I know fibre optic cables can be fragile. I'm not much more experienced, however, so there might be some hyper flexible, tough cable out there.
Great work
Great work. Also please make a video about the robot operating system (ros).
Your projects are very good. Can you make a video on your 3D printer
I wonder if it would be a good idea to send signals on wires made of stainless steel or carbon fiber or something that has higher resistance than copper. Might reduce induction effects. And since there's basically no power needed, it shouldn't result in enough heating/energy waste to matter.
You should make an autonomous or robotic RC zord
You could do a video to upgrade your droids robots using a omniwheel system to drive them, so you would have full range in 2 axis to control
I wonder about a shock absorber , you could measure the linear delta in the compliance . keep up the good work. looks awesome. id love to do smaller version I could follow along and make myself
WOOOOO! HOT DAMN!
What if you used transparent filament for the feet and use an optical sensor in the feet to see if contact is made with the ground?
I work professionally with load cells and strain gages. This is not how to use those load cells. It'll cause sensitivity to things you don't want. I'm glad you came to ditching the HX711, it's literal garbage. The way that load cell is setup is kind of like a parallelogram but with contraflexure (Thin element bends at both ends, one causing compression and one causing tension). That cut out has 4 gages in a wheatstone bridge on one side where one half of either element (top or bottom) is in compression and one half in tension due to how it flex's with the cutout. IF you mount it like that it's acting like a poor differential bending beam and you're constraining the contraflexure element. This becomes highly unpredictable and force calibration could change with bolt tension. So remove and remounting could cause huge changes in calibration. It will also decrease sensitivity and cause new sensitivities to things you likely aren't aware and potentially making it unstable.
Really, you might need a custom design because most load cells are 1 direction, normally perpendicular to the beam. However depending on what angle the leg is at, this might tell you only a partial story. A dual axis design would be better but few people deal with multi-axis load cells. I do, but I'm a rare case.
I'd start with RS485 for testing, then change everything to CAN
I don't think that's good advice; those protocols work quite differently.
Those leg motors all need to be moved in board to reduce the leg weight so they react much faster and accurate like the Boston dogs.
Yes that is true, if I wasn't so far into the project I'd rebuild the whole thing taking into account my dynamic leg R&D I did recently as well.
@@jamesbruton Maybe worth taking a page from Martin of Wintergaten's book? Angle grinder, anyone...? ;)
nice vid james verry cool
very nice video
thanks!
Load cell maybe workable, try using accelerometer chip? You will know exactly when you touch
Hi James
Instead of using CANbus have you considered ROS (robot operating system) ?
This would require a Raspberry Pi but is intended for this type of application.
CAN bus is a standard for data transmission (like serial), whereas ROS is for control of a system as a whole. I agree that using ROS would be a good step forward, however, and to top it off it can work with arduino so long as there is a Linux computer to send the appropriate messages.
Great project. You can use multiplex to deal with limited i2c addreas- might scale nicely to 4 legs. Pca9617 is handy to boost i2c signal. With your pwm signal did you put a terminating rc network in? Also a cheap mcu can act as an i2c slave.
Great video!! If everything else fails you can use your 2 bipeds as a cuadraped! (a la 2 person horse costume) :)
A robotic pantomime horse would be hilarious.
Maybe start making it walk using the ATAT walker style from Empire Strikes Back one foot at a time before going to a trot gait? baby steps!
I did in the past, but it has to lean sideways so much on each step
Amazing😍😍😍
Really Inspiring and knowledgeable
After seeing your video i do want to make cool stuff like you🤩🤩🤩
Very interesting , could you use some sort of band pass filter on the signal input to cancel the motor noise ? 👍🙂
what if you drove the signal into a resistor to get some current flowing through the PWM signal? Might mitigate some of the noise?
you should try making one of those little japanese boxing robots
If You need any help about CAN bus let me know ... differential signal will cancel your noise for sure . But maybe you can manage without it just with some smart wiring. Anyway there are plenty available libraries, you just need to choose some transceivers. Dunno if your uC have a CAN controller already implemented though ... if not you should use some other differential protocol.
mabye you could use fibre optic, I have a tesla coil and need to use fibre optic cables for the interupter because of all the interference the coil makes
Are you sure you are using the gait? There are six style gaits, and I think you are trying to use one that has forward motion already to help the balance. A tri-leg stance is not feet straight across but two wide apart and one in middle.
CAN bus! Check out CANopen used in robotics. I found it difficult to find any standards for it though. it looks like all messages are home grown. or you have to purchase message stacks. J1939 is a standard in wheeled vehicles but not very good for your application. Look forward to seeing what you come up with as far as the CAN buss!
What if you send the signal with a twisted pair of wires, one sending the signal, and the other sending the reference "ground" voltage, then you subtract the "ground" voltage from the signal to filter out the interference?
Yes CAN bus is a differential signal which does just that
Concerning the shielded cable, did you ground both sides of the shield? Because otherwise the shield won't work
He knows
This is not always true. Grounding one end should suffice. If you ground both ends of the shield but the ground at each have a different potential it can cause all sorts of problems.
@@FizzicksDude good point, but as far as I recall it, connectiong one end to ground is for analog signals and two ends is for digital signals
@@madome3117 Fair point! Another comment below seems to confirm it was the exposed board causing the trouble. Cheers!
The noise may not be in your shielded cable but from the load cell itself? Considering the level of amplification involved I would look at that end first.