I was wondering if anyone has implemented torque/force feedback? From what I’ve read Dynamixel Servo torque feedback measures current load and isn’t reliable beyond telling the robot it’s collided with something solid.
Are there any methods out there that would be sensitive and accurate enough to allow the motor to follow a near pendulum like movement by the attached limb? Such a method could be useful in developing walking strategies built around passive dynamics.
Another area it could be useful is if the robots ever use parallel actuators, or bi-articulated actuators. Telling a bunch of actuators to simply offer zero resistance to a connected limb movement would help controlling them.
This is about the best I found. Does anyone know of something better? Preferably cheap!
Torque feedback in older Dynamixel servos is even worse than that, it’s just the value of the PWM that the MCU uses to drive the motor.
The very new Dynamixel XM-430 have actual current sensing and it’s already much much better, but still suffers from the inertia and friction of the gearbox (plus these are pretty expensive models).
Funny thing is I have been looking very recently into that question, and came up with a very similar solution to this guy’s - although my version would be basically to measure the deflection of the 3D printed structural parts of the robot, and a magnetic position sensor instead of just a hall effect sensor. I wish I had see this earlier ^^
Most techniques are based on variations of “position sensor + spring”.
I have looked at and dissmissed: strain gage, FSR, electrogoniometer + spring, optical fiber goniometer + spring, measure of the resistance of compressed conductive foam.
One that looked cool I think was Led + photosensor + spring, but ultimately looks more complex than hall effect sensor + spring.
The MIT cheetah measures back-EMF in its custom brushless motors to sense force, but it kind of requires changing the whole motor paradigm, not the easiest thing to do. One guy at MIT then started work on a smaller version, see “hobbyking cheetah”.
With (much) more money on the table, and if you could work with sensing 3D force at the contact point instead, Optoforce is pretty awesome. ETH Zurich’s StarlETH use one on each foot.
Thanks for that reply Xevel. A lot for me to look into.
I’d also thought about the LED/photoresistor idea, but there was always the issue of outside light messing with the results. Magnetism is probably going to be more reliable.
I’m also thinking of using 3D printing. Probably SLS nylon. The main issue is likely to be inconsistency between sensors, due to the inconsistency between individual 3D prints. Each individual sensor should be reliable on it’s own.
I suppose I’ll just have to design a part and test it.
XL are like AX, they only have PWM value. I’m not entirely sure about the bigger MX, but IIRC they have the same problems.
In Wolftronix’ video, he shows 4 versions of the force sensing servo. The first two have the problem of being incapable of transmitting important torque, as they have a weakened mechanical structure.
The next two are very interesting though for an application in a walking robot, as they keep the servo lever intact.
Another thing to watch out for is torsion. If the method used to measure the deflection is sensitive to other changes like a torsion or flextion in a different direction, it would ruin the results.
One way to get rid of that problem I was thinking about but could not crack yet would be to find a way to make the position sensor truely sensitive to only one direction…
I was leaning towards SLS parts too - since it is extremely robust, flexible but not too much, capable of the craziest shapes, and cheap. I imagine using that occasion to make a cool looking robot at the same time as technically interesting.
I expect that the material will have some hysteresis problems that will have to be accounted for in software though. The alternative would be using much stiffer material and actual metal springs, to get less unwanted flexions and a behavior closer to the ideal model.
The idea is to use a pressure sensor in a silicone blob. Cheap and potentially awesome if done well, and they released all the instructions online.
Nowadays we can get pretty accurate i2c pressure sensors in insanely small packages and for less than 3 euros in quantity one (lps22hb, 2x2mm). Though not easily hand-solderable, it’s not too much of a problem with a stencil and a reflow oven (i.e. the cheapest convection oven I could find on amazon in my case ^^).
It has a wider range, better sensitivity, less noise than the one they used back in the day, for a similar sampling frequency.
Maybe an updated version could be used in a joint directly, with a 2-part structural piece joined by silicon, with the sensor embedded in the silicon…
Or it could be used to remake their examples as end effector/sensing surface with a better spacial resolution.