4 ms·

Do you think there's any chance of direct drive generating enough torque for a roughly human sized robotic arm with human like performance (say to lift a 20 kg

Do you think there's any chance of direct drive generating enough torque for a roughly human sized robotic arm with human like performance (say to lift a 20 kg load) ? The last time I looked into this it didn't look like that much torque was available in a small form factor. I found it rather surprising because larger scale industrial robot arms appear to far exceed any biological arm in both strength and speed.

macrolime 6y ago
There's at least some with 8kg payload https://www.halodi.com/ever3 https://www.halodi.com/ever3
thefounder 6y ago
I don't think a robot needs to lift more than 1kg or even less to be game changer. It just needs to be accurate and smart enough so that it can use different tools(i.e push different buttons) that do the heavy lifting.
- tgflynn 6y ago
  If all you want to do is push buttons there's probably a much simpler and cheaper way to accomplish what you want electronically.
  - Teever 6y ago
    I think we're talking about integrating robots into an environment that is designed for humans. As such robots pushing buttons is a good thing to reach for.
    - tgflynn 6y ago
      I still think that application is a dead end. Anything that has buttons on it these days (and increasingly so in the future) is likely to also have some kind of wifi or other remote and/or programmable interface so control of that machine really isn't something you need or want a robot to do. I'm not sure exactly what the parent had in mind but consider the problem of washing dishes. A robot that can load a dishwasher has value, a robot that can only press the start button not so much.
      - thefounder 6y ago
        Pressing buttons was just a figure of speach. It could use wireless commands if course. My point is that we need a smart robot to supervise and "use" various devices(i.e the dishwasher). It doesn't need to lift heavy things or do many tasks himself because it could use dumb tools/machines to do that just like humans use forklifts. That's why I'm saying being smart/ingeligent is more important than how much it can lift or its battery autonomy.
jamiek88 6y ago
Is it really just our arm that lifts that load though? Plenty of core muscles involved.
- tgflynn 6y ago
  That may be but if you want an equivalent robot arm driven by an ordinary motor you still need the torque. Otherwise you would need a completely different system of actuation. I've heard of "artificial muscle" systems but I'm not sure if they're sufficiently developed yet.
  - candiodari 6y ago
    It's difficult to see, but the arm and hand are as strong as they are because the "motors" are one step up. The motors for the hand are (predominantly) in the lower arm. The motors for the lower arm are in the upper arm, pulling on ropes connected to the lower arm (and significantly bigger than the lower arm). And the muscles doing the full arm are what's covering the front and back of your upper torso. This means that the force of your fingers is generated by a muscle roughly the size of a cola bottle, the force of your lower arm is generated by between 2 and 3 stacked cans of cola, and the force of your entire arm is generated by a muscle the size of a decent shoebox. So while muscles are pretty good, they're worse than our best motors. Or at least, if you can provide them with the power they need they are. And you think this is not the case because your body hides pretty sizeable muscles in your torso where everybody thinks you have almost only organs and fat, when in fact about half is muscle.
    - tgflynn 6y ago
      So are you saying that using cables and motor driven pulleys is a better solution ? Most robotic arms I've seen appear to have a motor located at the elbow joint that drives the equivalent of the lower arm either directly or through a gear box. If cables and pulleys give better torque without loss of speed why aren't they in more widespread use in robotic arms ?
      - 6y ago
Animats 6y ago
In industrial robot arms, the problem is not having enough power to lift the load. It's making the arm rigid enough that positioning the actuators will put the end at the desired point. Most industrial robot arms are position-controlled, usually rather blindly. Customers want repeatable motion to tight tolerances, like a machine tool. The mechanics are overbuilt and under-loaded to achieve that. I've seen direct drive SCARA robots with a big pancake direct drive motor on top. Those are used for vertical assembly, for designs where a straight-down move puts the part in place. The arm moves in X and Y, then there's a vertical piston-like actuator that moves in Z, and a rotation axis around Z.
- theptip 6y ago
  Anyone using machine vision to try to work around the positioning problem? Seems like working solely from proprioception is way harder than proprioception plus visual feedback.
  - Animats 6y ago
    Yes, there was a robot arm on HN about two weeks ago which had a QR code target on the end effector. Something that doesn't seem to have caught on is putting tiny cell phone cameras in the end effector, so you can fine-tune positioning.
    - tgflynn 6y ago
      The parent comment makes a good point. If accurate positioning is an issue it seems like you would want a closed loop control system that measures and optimizes for that. The position measurement itself could be made using machine vision or perhaps more precisely using lidar or some other technique. I've even wondered if that might allow you to build a less rigid and hence less expensive CNC mill.
- candiodari 6y ago
  There is lots of research pointing out that the problem may in fact be trying to do that at all. A significant issue with this is that humans can't do precision positioning of limbs in free space, yet can carry out all the tasks people want to use these robots for. When you see surgeons, who have loads of theory about precision movements you will quickly find that they rely on feedback rather precision for their movements. For example, they cut along the "linea alba", not by knowing where it is and positioning the knife exactly, but by simply slicing down starting at a good enough position, knowing the knife will follow the linea alba because it's attracted to it. Which is a fancy way of saying that it's a force feedback problem, not a positioning problem.
jnxx 6y ago
It is also that as soon as you have a rigid robot with that force, speed, and without a steel fence as separation near a human, you have a safety problem. If a human family member holds something like a baseball bat, or a butcher's knife within arms length of you, you can feel safe around them, and be sure they control their motions. Contrarily, having an industrial robot with something like a screwdriver at the height of your head and within a meter of distance is just scary.
- tgflynn 6y ago
  I'm not thinking of something as dumb as an industrial robot. You would want a safety mechanism that detects nearby humans (say in the same room) and switches to low power mode if any are present. That way you can get work done at a reasonable speed when the robot is alone with very little risk of hurting anyone. Yes, there are some safety issues to take into account with a household robot but they pale in comparison to those of a self-driving car whose whole reason for existence is to pilot a > 1 ton metal object at high speeds.

macrolime 6y ago

There's at least some with 8kg payload https://www.halodi.com/ever3 https://www.halodi.com/ever3

thefounder 6y ago

I don't think a robot needs to lift more than 1kg or even less to be game changer. It just needs to be accurate and smart enough so that it can use different tools(i.e push different buttons) that do the heavy lifting.

tgflynn 6y ago
If all you want to do is push buttons there's probably a much simpler and cheaper way to accomplish what you want electronically.
- Teever 6y ago
  I think we're talking about integrating robots into an environment that is designed for humans. As such robots pushing buttons is a good thing to reach for.
  - tgflynn 6y ago
    I still think that application is a dead end. Anything that has buttons on it these days (and increasingly so in the future) is likely to also have some kind of wifi or other remote and/or programmable interface so control of that machine really isn't something you need or want a robot to do. I'm not sure exactly what the parent had in mind but consider the problem of washing dishes. A robot that can load a dishwasher has value, a robot that can only press the start button not so much.
    - thefounder 6y ago
      Pressing buttons was just a figure of speach. It could use wireless commands if course. My point is that we need a smart robot to supervise and "use" various devices(i.e the dishwasher). It doesn't need to lift heavy things or do many tasks himself because it could use dumb tools/machines to do that just like humans use forklifts. That's why I'm saying being smart/ingeligent is more important than how much it can lift or its battery autonomy.

jamiek88 6y ago

Is it really just our arm that lifts that load though? Plenty of core muscles involved.

tgflynn 6y ago
That may be but if you want an equivalent robot arm driven by an ordinary motor you still need the torque. Otherwise you would need a completely different system of actuation. I've heard of "artificial muscle" systems but I'm not sure if they're sufficiently developed yet.
- candiodari 6y ago
  It's difficult to see, but the arm and hand are as strong as they are because the "motors" are one step up. The motors for the hand are (predominantly) in the lower arm. The motors for the lower arm are in the upper arm, pulling on ropes connected to the lower arm (and significantly bigger than the lower arm). And the muscles doing the full arm are what's covering the front and back of your upper torso. This means that the force of your fingers is generated by a muscle roughly the size of a cola bottle, the force of your lower arm is generated by between 2 and 3 stacked cans of cola, and the force of your entire arm is generated by a muscle the size of a decent shoebox. So while muscles are pretty good, they're worse than our best motors. Or at least, if you can provide them with the power they need they are. And you think this is not the case because your body hides pretty sizeable muscles in your torso where everybody thinks you have almost only organs and fat, when in fact about half is muscle.
  - tgflynn 6y ago
    So are you saying that using cables and motor driven pulleys is a better solution ? Most robotic arms I've seen appear to have a motor located at the elbow joint that drives the equivalent of the lower arm either directly or through a gear box. If cables and pulleys give better torque without loss of speed why aren't they in more widespread use in robotic arms ?
    - 6y ago

Animats 6y ago

In industrial robot arms, the problem is not having enough power to lift the load. It's making the arm rigid enough that positioning the actuators will put the end at the desired point. Most industrial robot arms are position-controlled, usually rather blindly. Customers want repeatable motion to tight tolerances, like a machine tool. The mechanics are overbuilt and under-loaded to achieve that. I've seen direct drive SCARA robots with a big pancake direct drive motor on top. Those are used for vertical assembly, for designs where a straight-down move puts the part in place. The arm moves in X and Y, then there's a vertical piston-like actuator that moves in Z, and a rotation axis around Z.

theptip 6y ago
Anyone using machine vision to try to work around the positioning problem? Seems like working solely from proprioception is way harder than proprioception plus visual feedback.
- Animats 6y ago
  Yes, there was a robot arm on HN about two weeks ago which had a QR code target on the end effector. Something that doesn't seem to have caught on is putting tiny cell phone cameras in the end effector, so you can fine-tune positioning.
  - tgflynn 6y ago
    The parent comment makes a good point. If accurate positioning is an issue it seems like you would want a closed loop control system that measures and optimizes for that. The position measurement itself could be made using machine vision or perhaps more precisely using lidar or some other technique. I've even wondered if that might allow you to build a less rigid and hence less expensive CNC mill.
candiodari 6y ago
There is lots of research pointing out that the problem may in fact be trying to do that at all. A significant issue with this is that humans can't do precision positioning of limbs in free space, yet can carry out all the tasks people want to use these robots for. When you see surgeons, who have loads of theory about precision movements you will quickly find that they rely on feedback rather precision for their movements. For example, they cut along the "linea alba", not by knowing where it is and positioning the knife exactly, but by simply slicing down starting at a good enough position, knowing the knife will follow the linea alba because it's attracted to it. Which is a fancy way of saying that it's a force feedback problem, not a positioning problem.

jnxx 6y ago

It is also that as soon as you have a rigid robot with that force, speed, and without a steel fence as separation near a human, you have a safety problem. If a human family member holds something like a baseball bat, or a butcher's knife within arms length of you, you can feel safe around them, and be sure they control their motions. Contrarily, having an industrial robot with something like a screwdriver at the height of your head and within a meter of distance is just scary.

tgflynn 6y ago
I'm not thinking of something as dumb as an industrial robot. You would want a safety mechanism that detects nearby humans (say in the same room) and switches to low power mode if any are present. That way you can get work done at a reasonable speed when the robot is alone with very little risk of hurting anyone. Yes, there are some safety issues to take into account with a household robot but they pale in comparison to those of a self-driving car whose whole reason for existence is to pilot a > 1 ton metal object at high speeds.