hopping humanoids, continued

    Regenerative servomotor controllers do exist, although generally
for larger motors.  Now that we have ultracapacitors, that could
be scaled down.

    But that's not the problem.  The problem is building a low-weight
drive system which can usefully absorb the shock of landing.
Air and hydraulic cylinders can do that well, but it's hard to
do with electric drive.  Linear motors remain too heavy for
their power (although Aura, before they went bankrupt, was
at 10:1 power/weight ratio).  Leadscrew drives usually aren't
back-driveable, although maybe with a low-pitch ball screw
you could make it work.  Schemes with strings and belts are
usually troublesome under shock loads.   Series elastic
actuators, with a stiff spring after the motor, can take shock
loads but usually can't transmit them usefully back through
the leadscrew.  Some biologically oriented "robot muscle" approaches
are promising but don't really work yet. Motor-spring-clutch systems
tend to be too complex mechanically byt have been tried.


    "AC synchronous" and "brushless DC" motors are the same thing,
and yes, those are widely used in robotics.

                John Nagle

John, I'm not sure what you mean by this... With a series elastic actuator,
the force sensor gives feedback, and the motor controlling the leadscrew
can quickly back off the actuator to react to the load.

I see SEA's as the perfect actuator for biped control. For jumping, I see
the robot keeping the legs almost straight, like a person would, with the
force applied to the actuators almost zero. The instant the legs start to
compress, the force-driven actuators would start to quickly increase the
force applied to each actuator, until full compression is reached.

The key is that while the force desired for each actuator is low, the
leadscrew would be backing off quickly. As the desired force increases, the
leadscrew would slow down, until the force applied is at a maximum, and
hopefully the springs are not quite fully compressed.

Later,
Jon

--------------------------------------------------------------
   Jon Hylands      Jon@huv.com      http://www.huv.com/jon

  Project: Micro Raptor (Small Biped Velociraptor Robot)
           http://www.huv.com/blog

     The original poster was asking about energy storage.  Series elastic
actuators don't recover much energy, because the spring travel is small
relative to the entire actuator travel.

     Humans recover about 70-80% of the energy in running through
elastic storage.  Cheetahs reach about 90%.  Without energy recovery,
you need bigger motors and power sources to build a dynamic running
machine.  That's a design constraint, but you can build a running
machine that has no elastic energy recovery.  The Honda ASIMO doesn't.

     Trooby, the closest thing to what Hylands is talking about, does
have elastic energy recovery.

                John Nagle

Ahh, yes, good point.


Well, I think Asimo (and most of the other balancing bipeds) don't actually
run, they shuffle quickly. All of the bipeds that I have seen that run do
so attached to a pole that keeps them balancing. But energy recovery is
something that is interesting to think about. I was under the impression
that humans just didn't apply much energy in the first place to walk or
run.


Troody was the inspiration, at least physically, for the robot I want to
build. The way it walks, in my mind, was incredibly inspiring.

Later,
Jon

--------------------------------------------------------------
   Jon Hylands      Jon@huv.com      http://www.huv.com/jon

  Project: Micro Raptor (Small Biped Velociraptor Robot)
           http://www.huv.com/blog

No, Asimo really can run, with no external support.  Well, OK, it's more
of a jog than a run, but it does get both feet off the ground briefly.

It's not the only one, either -- the first free-running biped was the
much smaller Qrio, in 2003 IIRC.

Here's a useful story:
  <http://news.bbc.co.uk/1/hi/technology/4098201.stm>

Best,
- Joe

I've seen Qrio run, and I still don't really consider that running, both
feet off the ground or not.

Probably I'm just being picky, but with both feet constantly level and
moving straight forwards and backwards, it looks more like a shuffle than a
run.

Later,
Jon

--------------------------------------------------------------
   Jon Hylands      Jon@huv.com      http://www.huv.com/jon

  Project: Micro Raptor (Small Biped Velociraptor Robot)
           http://www.huv.com/blog

Asimo shuffles because it can't really run. Running requires leaping
into the air such that all legs are off the ground, at least part of
the time. None of the common japanese bipeds, like Asimo or the Robo-1
type, can do this. They are motor-powered 100% of the time, and the
motors used just wouldn't allow the feet to all leave the ground
simultaneously.

Animals solved this problem, as John.N noted, by storing and releasing
much of the required energy for jumping in the elastic elements of the
legs. There is a physiological limit for how fast an animal can go,
given a particular gait. To do a fast trot, for instance, the legs have
to move very fast, and are limited by how fast the muscles can flex and
relax. I have exactly the same problem with my servo-driven hexapod.
The servos only slew so fast.

To go faster than the fastest trot, the animal has to change gait to a
gallop, for instance, where the animal pushes its body off the ground
from both rear legs simultaneously. It gains speed mainly by greatly
lengthening its stride, rather than by moving its muscles faster. It
also pays a large energy cost for this, because trotting animals can
trot for hours, while galloping animals, especially cheetahs, can only
keep this up for seconds or a minute or two. See how hard the cheetah
breathes after its sprint.

Humans have exactly the same problems. They can walk for hours, but can
only run very fast for a few seconds. During a 100-yard dash, the legs
are moving as fast as possible and the stride is extremely long, and
the muscles run out of energy quickly. During a 10-KM run, the legs
move much slower and the stride is also shorter.

Walking humans don't do as much energy store-release because their legs
are basically held straight, with little flex, during walking. This
gait is described using the "inverted-pendulum" model. During running
however, the legs flex heavily as they come down to the ground, and
straighten during the power stroke. This gait is described using the
"spring-mass" model. A lot of this is described here ...

http://polypedal.berkeley.edu/twiki/bin/view/PolyPEDAL/PolypedalPublications

R.J. Full and C.T. Farley, 2000, "Musculoskeletal Dynamics in Rhythmic
Systems: a Comparative Approach to Legged Locomotion", in Biomechanics
and Neurocontrol of Posture, Springer.

I'm curious why you say this. Are you implying that a motor-drive servo
can't make a robot leave the ground?

I've seen videos of a Robo-One robot that is jumping an inch or so in the
air, with both feet.

I plan to experiement with this once I get my Bioloid kit. I plan to
machine custom feet for the robot, which will include curved heel and toe
sections (with rubber grip) to allow push-off and landing. It will be
interesting to see what happens...

Later,
Jon

--------------------------------------------------------------
   Jon Hylands      Jon@huv.com      http://www.huv.com/jon

  Project: Micro Raptor (Small Biped Velociraptor Robot)
           http://www.huv.com/blog

That's two of you guys behind the times now.  :)  Qrio ran in 2003, and
Asimo has been doing so since late 2004.  There are probably other
running bipeds by now, but it's no longer newsworthy.  This is using
your definition above (which is the standard one) of both feet being off
the ground for some part of the stride.

Though I agree with Jon's point that it doesn't look like a run as we
informally use the term -- though to me it's clearly a jog, not a
shuffle.  (A shuffle to me is a gait where you always have at least one
foot on the ground, and in fact have both feet on the ground most of the
time.)  But technically speaking, these bots can run.


Right.  And like the others in this thread, I too am very interested in
how robots can apply these energy storage and release principles for
more efficient locomotion.

Best,
- Joe

In humans, I believe that the distinction between a jog and a run
is whether most of the rebound energy (stored on each impact) is
used in the takeoff. So by your description, jog is the right word.

     Asimo actually does have a moment of suspension at the run,
but you have to look at a video taken from foot level and step
through it frame by frame to see the 1cm of ground clearance.

     Some of the little Japanese hobbyist machines in the $1000
category have powerful enough actuators to run, but don't have the
balance software to do it.  See

     http://www.robots-dreams.com/2006/12/greased_lightni.html#more

     The off the shelf hardware from Japan has become quite good.
As soon as those things get 6DOF inertial sensing, the hardware
will be ready.

     These robots don't have energy recovery in the legs, but for
smaller robots, that's less of an issue.  Energy recovery is
a cube/square scaling issue.  (Mass goes up as the cube of
length, but strength goes up as the square.  That's why
insects have skinny legs and elephants have huge ones.)

                John Nagle

To answer John.H and Joe, etc, I've seen the videos of the japanese
robo-1 that took a hop., and heard that Asimo can get both feet 1 whole
cm off the ground at the same time, but this is light years behind
actual running in dogs and humans.

The real difference is, as mentioned in John.N's post here, that real
animals use significant amounts of energy storage+release in elastic
elements in their legs, and also bending spine movements to some lesser
degree. This is what needs be incorporated into bots to get a real run.


The digital servos may be just about powerful enough to do it, I'm not
sure, but it's much more efficient to use springs/springiness, which
translates into getting your extra needed energy from gravity. As I
mentioned last time, look up the spring-mass model.

Well, yeah.  When the robot uprising comes and your biped robot is
trying to kill you, the best way to get away from it is to walk away at
a leisurely pace.  :)

Nobody's arguing that today's robots are anywhere remotely close to the
performance of live animals.  My point was only that we really should
stop saying things like "robots can't run" once it's been demonstrated
that they can.  But you could continue to say that they can't run
fast/well/impressively/whatever.


Right.  But, returning to the original poster's question, it sounds like
electrical energy recovery and storage is problematic, at least with
standard servos.  Springs don't work very well either for a robot that
you want to also be able to stand still (though I've seen them used well
on robots that ONLY bounce/run).  It could be that the right actuators,
or combination of actuators and passive parts, have not yet been
developed.

Best,
- Joe

HA! You can't scare me. I read the book about the robot uprising. There
are many ways to survive a killer robot. Regards Asimo, all I'll have
to do is extend a rope "2" meager cm above the ground, and it'll have
to stop and think for about 3 minutes, during which time, I'll make my
getaway.




I've clearly made the distinction several times about what it takes to
perform real running. Walking fast or trotting aren't the same thing.
The biomechanics of the different gaits are very different, and it's
wrapped up in the idea of the inverted-pendulum versus spring-mass
models, and energy utilization, that I've mentioned several times
already.


type, and to get real-running will probably require adding in the
elastic part. I'm sure Asimo's designers understand this too.

And once this is done, then the batteries will last much longer. As
mentioned, the transition from walk-trot gaits to run-gallop gatits
occurs at a specific speed range for animals because the energy
requirements to walk-trot any faster are unrealistic. The muscles just
can't handle. Likewise there is an upper limit to to how fast humans
can move in those fast-walking contests [whatever they're called],
where they wiggle their hips like ducks.

You should really go read some of Robert Full's papers that I cited.
He's the Elvis of locomotion.

Agreed.  But there is a standard, widely accepted definition of running,
and it's simply that both feet leave the ground for some part of the
cycle.  Using words in standard ways aids communication.


You really should quit assuming that I haven't.  :)

Best,
- Joe

So hopping is running, then.



You might look up the differences in gait biomechanics, vis-aviz foot
action and energy untilization. How does the foot pattern differ
between walk/trot and gallop. Why does a transition occur. Why doesn't
an animal gallop at slow speed. Why can't an animal trot as fast as it
can gallop. Why does the inverted-pendulum apply better in one case,
while the spring-mass model is better for the other.

    There have been a few attempts, but they're mechanically clunky,
like schemes where a string winds around a pulley with a corner cut,
so that when the string is fully wound, a bit more motor rotation
pushes the string off the pulley and the string releases all at once.
Some people at Stanford tried that.  Sort of worked, but not too well.

    The best known solutions for larger robots are either hydraulic or
pneumatic.  A muscle can be thought of as a spring with an adjustable
spring constant and an adjustable zero point.  You can get that effect
with a double-ended pneumatic cylinder - high pressure on both sides
makes a stiff muscle, low pressure makes a limp one.  Unequal pressure
changes the zero point.  That's been tried, too.  Works fine, but you
need to carry an air compressor or have external hoses.

    The same trick can be played with hydraulic cylinders and accumulators,
and that's been used in Boston Robotics' Big Dog, which is pony-sized.

     For desktop robots, high end servos, maybe with some stiff
springyness in the legs to absorb shock loads, seems to be the way to
go for now.  Right now, we need better control algorithms, not higher
efficiency.  Energy recovery will improve battery life, but that's not
the current top problem.

                John Nagle
                Animats

What about pair of magnetic clutches to disconnect the servo
transmission and hook it up as a generator ? Sounds clunky and
expensive ..
Also, what about reversible tranmission types like planetary or
harmonic drive ?

-kert

BioLoid kit is around $1000, 6DOF IMU costs $300 nowadays, 5DOF $100,
and ample portable computing power next to nothing, in form of 32-bit
micros.
http://www.sparkfun.com/commerce/categories.php?cPath#_85

BTW, why only Japanese/Eastern, whats holding back the US ?

-kert