Loading ...
Sorry, an error occurred while loading the content.

Re: [SeattleRobotics] Re: Pneumatics parts for robots

Expand Messages
  • David Buckley
    ... Sort of like animal life ... From: Kevin Ross To: SeattleRobotics@yahoogroups.com Sent: Wednesday, May 05, 2010 7:30 AM Subject: Re: [SeattleRobotics] Re:
    Message 1 of 27 , May 5, 2010
    • 0 Attachment
      > your robot will be hissing and fidgeting the entire time
      Sort of like animal life
       
      ----- Original Message -----
      Sent: Wednesday, May 05, 2010 7:30 AM
      Subject: Re: [SeattleRobotics] Re: Pneumatics parts for robots

       

      Honestly, pneumatics will be the bane of your existence if you are planning for fine control. Air compresses like a spring. You will be forever making small adjustments, meaning your robot will be hissing and fidgeting the entire time. You will consume a massive amount of air and power.
       
      It could be an interesting experiment to attempt to use the fluctuations in cylinder pressures as a sensory feedback. I see potential with that.

      Sent: Tuesday, May 04, 2010 9:34 PM
      Subject: Re: [SeattleRobotics] Re: Pneumatics parts for robots


      Any neural model is going to depend heavily on rich, real-time feedback.  My model is very good at temporal correlations- -that is, it learns spatial/temporal patterns and can judge whether or not the current experience is more or less normal from past experiences, in context.  Taking time to change pressure balances in a cylinder is probably a good thing because it leaves a larger temporal signature to correlate with whatever else might overlap or be near is time-proximity.  For example, how movement in one joint under a given context might change pressures in other joints and/or balance (via gyro).

      I don't hardwire gaits. Instead, I dynamically form inhibitory receptors between mutually exclusive neural pathways.  

      Matthew

      On Tue, May 4, 2010 at 9:47 AM, Peter Balch <peterbalch@btintern et.com> wrote:
       

      > Mostly, I need cheap pressure gauges with digital output (not LCD
      > displays)
      > for to plug into a microcontroller.

      I've absolutely no feel for what pneumatics is like in a closed loop at
      speed. With an electric motor it's easy (in theory) to calculate the
      momentum and therefore calculate the PID parameters. But with pneumatics
      (i.e. a compressible fluid) there must be far larger problems of overshoot
      because of the delay between a valve changing and the pressure changing or
      the joint coming to rest.

      Do you need an second-derivative or second-integral or feedforward term or
      do you add the current pressure into the controller math somehow.

      Is that why you need lots of pressure gauges?

      Peter


    • David Buckley
      If you are going to use cylinders then you really need to use glass cylinders and pistons with no seals as sarcos originaly did. ... From: dcwjobs2004 To:
      Message 2 of 27 , May 5, 2010
      • 0 Attachment
        If you are going to use cylinders then you really need to use glass cylinders and pistons with no seals as sarcos originaly did.
         
        ----- Original Message -----
        Sent: Wednesday, May 05, 2010 7:31 PM
        Subject: [SeattleRobotics] Re: Pneumatics parts for robots

         



        I agree.

        Air based pneumatics can be good for open-loop, on-off, push/pull applications at rates of up to 1 Hz and with work to maybe 5 Hz. This assumes a pressure of 30-100 psi. Pneumatics can be fun for making open loop Rube Goldberg machines, like mechanical merrygoround organs.

        Closed loop control of pmenumatics is tough. Bandwidths of 1/5 Hz are OK with a little work. You might just be able to get to 1 Hz with a *lot* of work and side-effects.

        All kinds of things come into play.

        1. Air propagates at ~1 msec/foot; electricity at 1 nsec/foot.
        2. Small tubing has lots of resistance to fast air flow.
        3. The air in the cylinder works like a capacitor.
        4. So at low velocities ("DC"), things look fine, but the response time to change is slow, even in open loop.

        5. The air "capacitor" size changes with the position of the piston, a dynamic, non-linear parameter change.

        6. The tubing between valve and cylinder also has "capacitance" .

        Pneumatics means air. If we substitute oil or water for the air, we have hydraulics, which is much better for speed and power. A big part of this is that liquid does not compress much. You get little "capacitive" effect in the tubing and cylinders. Aircraft has used hydralics with bandwidths in the 100 Hz range using 1500-3000 psi.

        OTOH, hydraulics leak. And leaky hudralics are unpopular. Don't leak fluid on the living room rug. They are also much more expensive than pneumatics, partly because they are 15-30 times more powerful. I notice that aircraft is moving from hydralics to electric motors for moving things, both military and commercial.

        Pnuematics can be cheap and handy for certain things. Just do not try to use them to close loops, such as to balance a robot.

        --- In SeattleRobotics@ yahoogroups. com, "Kevin Ross" <kevinro@... > wrote:
        >
        > Honestly, pneumatics will be the bane of your existence if you are planning for fine control. Air compresses like a spring. You will be forever making small adjustments, meaning your robot will be hissing and fidgeting the entire time. You will consume a massive amount of air and power.
        >
        > It could be an interesting experiment to attempt to use the fluctuations in cylinder pressures as a sensory feedback. I see potential with that.
        >
        >
        > From: Matthew Tedder
        > Sent: Tuesday, May 04, 2010 9:34 PM
        > To: SeattleRobotics@ yahoogroups. com
        > Subject: Re: [SeattleRobotics] Re: Pneumatics parts for robots
        >
        >
        >
        >
        >
        >
        >
        > Any neural model is going to depend heavily on rich, real-time feedback. My model is very good at temporal correlations- -that is, it learns spatial/temporal patterns and can judge whether or not the current experience is more or less normal from past experiences, in context. Taking time to change pressure balances in a cylinder is probably a good thing because it leaves a larger temporal signature to correlate with whatever else might overlap or be near is time-proximity. For example, how movement in one joint under a given context might change pressures in other joints and/or balance (via gyro).
        >
        >
        > I don't hardwire gaits. Instead, I dynamically form inhibitory receptors between mutually exclusive neural pathways.
        >
        >
        > Matthew
        >
        >
        > On Tue, May 4, 2010 at 9:47 AM, Peter Balch <peterbalch@ ...> wrote:
        >
        >
        >
        > > Mostly, I need cheap pressure gauges with digital output (not LCD
        > > displays)
        > > for to plug into a microcontroller.
        >
        >
        > I've absolutely no feel for what pneumatics is like in a closed loop at
        > speed. With an electric motor it's easy (in theory) to calculate the
        > momentum and therefore calculate the PID parameters. But with pneumatics
        > (i.e. a compressible fluid) there must be far larger problems of overshoot
        > because of the delay between a valve changing and the pressure changing or
        > the joint coming to rest.
        >
        > Do you need an second-derivative or second-integral or feedforward term or
        > do you add the current pressure into the controller math somehow.
        >
        > Is that why you need lots of pressure gauges?
        >
        > Peter
        >

      • Matthew Tedder
        I think the tendencies of air compression will be of benefit to my neural mechanism. The time overlaps and bouncing between movements within and between
        Message 3 of 27 , May 7, 2010
        • 0 Attachment

          I think the tendencies of air compression will be of benefit to my neural mechanism.  The time overlaps and bouncing between movements within and between joints should help signature particular movement interaction types for abstraction.  I can see the issues with conventional neural nets, particularly all those that send individual analog values between neurons, instead of digital bits with frequency determining receptor contribution to the soma.  I calculate for short-term potentiation behind each receptor.  I also calculate for long-term potentiation (an equivalent of multiple receptors from the same axon).  And, I form and destroy both excitatory and inhibitory interconnections dynamically.  The more two pathways tend to excite in a mutually exclusive manner, the more they will inhibit each other.   It's exceptionally good at molding the spatial/temporal interaction patterns of experience.  The more normal the current experience is, as compared with all past experience (very richly considered), the stronger the representative pathway will excite.

          The other part of this, re-implementing my logic (currently in C++) to VHDL, has been very difficult for me.  My only effective work has been in the GNU VHDL compiler because the learning curve on Xilinx's tools.  I wish an electrical engineer handy.  I have no formal training beyond basic electronics and your typical computer architecture course in this sort of thing.

          There doesn't seem to be any alternative to Xilinx for actually programming my (or any) FPGA board... at least not that I know of.  I'd like something simpler.  Just downloading the Xilinx webkit took a very long time, installing even longer (due to downloading updates).  Did I choose all the right options in this process? I am really unsure of that.  I really very much wish I could just program my VHDL code into the FPGA over its USB connection and start playing with I/O pins.

          To think, I thought Visual Studio was huge, sluggish and buggy.

          Matthew

          On Wed, May 5, 2010 at 1:30 AM, Kevin Ross <kevinro@...> wrote:
           

          Honestly, pneumatics will be the bane of your existence if you are planning for fine control. Air compresses like a spring. You will be forever making small adjustments, meaning your robot will be hissing and fidgeting the entire time. You will consume a massive amount of air and power.
           
          It could be an interesting experiment to attempt to use the fluctuations in cylinder pressures as a sensory feedback. I see potential with that.

          Sent: Tuesday, May 04, 2010 9:34 PM
          Subject: Re: [SeattleRobotics] Re: Pneumatics parts for robots


          Any neural model is going to depend heavily on rich, real-time feedback.  My model is very good at temporal correlations--that is, it learns spatial/temporal patterns and can judge whether or not the current experience is more or less normal from past experiences, in context.  Taking time to change pressure balances in a cylinder is probably a good thing because it leaves a larger temporal signature to correlate with whatever else might overlap or be near is time-proximity.  For example, how movement in one joint under a given context might change pressures in other joints and/or balance (via gyro).

          I don't hardwire gaits. Instead, I dynamically form inhibitory receptors between mutually exclusive neural pathways.  

          Matthew

          On Tue, May 4, 2010 at 9:47 AM, Peter Balch <peterbalch@...> wrote:
           

          > Mostly, I need cheap pressure gauges with digital output (not LCD
          > displays)
          > for to plug into a microcontroller.

          I've absolutely no feel for what pneumatics is like in a closed loop at
          speed. With an electric motor it's easy (in theory) to calculate the
          momentum and therefore calculate the PID parameters. But with pneumatics
          (i.e. a compressible fluid) there must be far larger problems of overshoot
          because of the delay between a valve changing and the pressure changing or
          the joint coming to rest.

          Do you need an second-derivative or second-integral or feedforward term or
          do you add the current pressure into the controller math somehow.

          Is that why you need lots of pressure gauges?

          Peter



        Your message has been successfully submitted and would be delivered to recipients shortly.