I have a particular example robot that interests me:


See first image, the bigger robot on the left, in particular the shoulder pitch joint that would support the most weight. I'm curious because I know it's really hard to balance strength and precision in these types of robots, and want to know how close a hobbyist could get.

Would they be something similar to this: rotary tables w/ worm gears?


Looking for more actuation schemes to research, thanks!


Actuation is a complex subject that cannot be reduces to 1 component alone. It is the interaction of software electronics and mechanics. "Toy" servos tend to incorporate everything in one small package. Industrial robots have 3 or 4 separate components (each costing several hundred or a few thousand euros) instead.

To answer your question, worm gears are almost never used in the joints of industrial robots. Mostly cycloidal or planetary gears are used.

Transmission: Many hobbyist project use small "toy" servos which are, basically a DC motor, some simple position feedback system and a transmission (+ the controller). The transmission is a 1, 2 or sometimes 3 stage simple gear reducer. Due to its low quality it sometimes has large backlash, so the output of the transmission can move a bit (approx. 0.1 degrees, value varies on quality) even if the motor does not move. This leads to uncontrollable motions, vibrations due to jerk, etc. The best "toy" servos you can get, to my knowledge are the Dynamixel servos.

Industrial robots us professional gearboxes with backlash below 0.2 arc minutes. Price for something that you pointed out in the picture is probably around 1000-3000 EUR. Typical manufacturers: ZF, Wittenstein, Nabtesco, Bonfiglioli. Furthermore, the control algorithm applied backlash compensation to compensate these and further improve the precision.

Motor: As mentioned, the "toy" servos use DC motor in most cases. DC motor in general my not be cheaper then PMSM (what industrial robots use, see below) but due to bulk manufacturing in small sizes and cheaper motor drivers, overall in the small size range, they may be cheaper to use.

Industrial robots use Permanent Magnet Synchronous Motors. They are less expensive in general as DC Motors (also due to lower maintenance costs, since there is no brush to change regularly). Power ranges for smaller robots are a few hundred watts for larger Robots a few kilowatts. Price for such a motor varies from approx. 300 EUR to 2000 EUR for the high powered ones. Most of them include (optionally) resolver or encoders for position control. These have high precision (up to 4048 values per rotation with additional signal for in-between values). Typical manufacturers form for motors ABB, Siemens, Bosch, Parker. For best performance, encoders are integrated in the motors.

Motor Controller Toy servos use in most cases 1 Position control loop with PID control method. This is very simplistic and delivers bad performance.

Industrial motor controllers use cascaded control loops (each loop is a variant of PID) for position, velocity and torque. This increases significantly their performance. To increase the performance of the controllers, offset values are added to the control loops (known as feed forward control). This way the controllers cope easier with their task (e.g. if acceleration is needed, the fed forward controller applies an offset to the control output of the torque loop the make the reaction time faster, the weight of the robot is compensates as an offset value on the torque coming from a static model, etc.). Industrial motor controllers have the same manufacturers as motors. Usually it is best to use a matching manufacturer.

Another difference is how the reference signals are given to the motor controllers. This is "only software" so it seems not expensive, but a lot of reaseach is in this domain. THe question is how to plan trajectories between two points so that the motor controller can deliver its best perfromance. You can read about this here.

Al of thea measure above is to "balance strenght and precision" as you formulated. Servos in hobbyist application are nowhere near their counterparts in industrial robots in terms of build quality and performance. I am afraid there is no cheap solution to get the same performance. The cheapest way to improve the performance of hobby equipment it though software. Get a servo where the cascaded control is implementable (like the dynamixel ones) and implement trajectory generators (many robotics courses are available online, like the link I sent you) that plan a trajectory while limiting jerk. Fine tune the control loop of the servo for your robot (instead of accepting the factory default). Compensate anything than can be calculated, eg. the static torque needed to hold the robot still as a feed forward controller.

With the above measures you can at leas get the most out of the hobbyist hardware, which will already make a difference.

  • $\begingroup$ Thank you very much for your comprehensive answer. As I learned, commercial DC gearmotors and servos are garbage -- too much backlash. I've been prototyping with stepper motors (a type of PMSM?) and timing belt drives to create my actuators. I find the mechanical construction difficult but I think this approach has potential. What type of performance do you think I could reasonable get with this approach? I understand it's the same tech used in 3D printers, which is why I have been trying. I got bigger motors for belts for more torque. $\endgroup$
    – JDS
    Sep 8 '16 at 4:28
  • 1
    $\begingroup$ Belts are flexible and so they can be deformed. Care has to be taken when they are selected. Industrial robots also use belt transition in their wrist sometimes, but usually some form of compensation for their elongation is applied. 3D printers do not have high payload or high speed so belt deformation is negligible. This will not be the case for industrial robots. You can find high quality commertial DC motors with gears, but they are not cheap. Try Maxon motors form example. They specify backlash for their gearboxes. $\endgroup$
    – 50k4
    Sep 8 '16 at 10:51
  • 1
    $\begingroup$ Steppe motors do not have a position feedbacck. You just assume that they did thee step you commanded them to do. In some cases, if torques are high, they can slip (skip the step you commanded them to do). Unless you use an advanced control algorithm that detect/eliminates these slips you will have position errors that accumulate over time and you will have no feedback to correct these. $\endgroup$
    – 50k4
    Sep 8 '16 at 11:01
  • $\begingroup$ Thanks for the follow up. Regarding steppers, I usually get models w/ backshaft so I can mount rotary encoder for feedback. You are right, belts will elongate which is unfortunate. Perhaps I haven't used them enough, but so far my tests are going OK. In terms of gear motor backlash, would you be able to estimate an acceptable number - in arcmin - for a shoulder pitch joint at 1m length to achieve millimeter (+/- 1mm) precision and repeatability? E.g. I was considering buying a model with 15 arcmin backlash, not sure if that is good. Does backlash increase with use? $\endgroup$
    – JDS
    Sep 8 '16 at 16:47
  • $\begingroup$ Btw, my first goal is open-loop mechanical precision on a simple 1 meter long robot arm with 6 DoF (standard config as in the OP Foxbot link). $\endgroup$
    – JDS
    Sep 8 '16 at 16:49

A robot like the one you refer to probably has a mix of transmission types that increase in power-to-weight ratio and decrease in size and stiffness as you move outwards on the kinematic chain. The base joints likely have planetary or cycloidal gearing, with maybe some spur gears in the middle, and maybe a harmonic drive reducer at the end effector. I don't think worm gears are typically used with these sorts of revolute joints, unless there's maybe some special, application-specific tool on the end.

If you want to build a hobbyist robot that does some cool stuff, I'd stick with spur gears and planetary gears while you wade through the dozens of other difficulties in designing serial-link manipulators.

However, if you're looking to research interesting actuation schemes that maximize power-to-weight ratio and precision in a robotic joint, then harmonic drives are worth looking into. What you gain in compact size, power-to-weight ratio, and precision, you pay for in joint stiffness and friction losses, so your robot won't get snappy accelerations on the joints that use harmonic drives if that's something you're looking for. A serial chain that uses entirely harmonic drives will need to move pretty slowly to keep resonance under control. You can check out Rethink Robotics's Youtube videos to see what I'm talking about - I suspect they uses HDs on most if not all joints. Harmonic drives are also very difficult to assemble, since the gearing needs to be pre-strained, so they tend to cost a lot more.

Hope that answers your question. Good luck.

  • $\begingroup$ Thanks for your comment. Planetary and spur gearmotors - what I can afford commercially - have awful backlash. I've been having better results w/ steppers and belt drives, but tensioning belts is hard. Harmonic drives are not realistically affordable. The Series Elastic Actuators used by Baxter are both too complex and not precise enough for what I'm after unfortunately. $\endgroup$
    – JDS
    Sep 8 '16 at 4:23
  • 1
    $\begingroup$ FYI, The Kinova Robotics, Universal Robotics, and probably also Rethink's Sawyer arms use harmonic drives. The Rethink Baxter simply uses a spur gear train with a series compliant element. $\endgroup$
    – Ben
    Sep 9 '16 at 16:53

"Would they be something similar to this: rotary tables w/ worm gears?"

The used mechanisms would vary based on the size of the robotic arm and the application.

There would be plenty of gears used (spur gears, transmission belts, harmonic gears, etc.), depending on the size of the robotic arm and the application (big traditional industrial arm, cobot arm, small pick-n-place, etc.)

Concerning hobbyist versions, one of the main constraints is size (considering that the hobbyist is willing to pay for his robotic prototype), which would limit the design choices available.

How near hobbyist ones would get depends on what suppliers (hobbyist suppliers) would provide in the near future, because it is infeasible to custom make every single component.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.