Newsgroups: comp.robotics
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!newsxfer.itd.umich.edu!caen!msuinfo!harbinger.cc.monash.edu.au!news.uwa.edu.au!DIALix!metapro!bernie
From: bernie@metapro.DIALix.oz.au (Bernd Felsche)
Subject: Re: Two-legged robots
Message-ID: <CznDy2.A78@metapro.DIALix.oz.au>
Organization: MetaPro Systems, Perth, Western Australia
References: <hallumdCyz6GG.Io3@netcom.com> <39voqd$4r5@mutton.csv.warwick.ac.uk> <3aihsg$au2@seralph9.essex.ac.uk> <3aog72$lme@handler.Eng.Sun.COM>
Date: Tue, 22 Nov 1994 02:37:13 GMT
Lines: 56

In <3aog72$lme@handler.Eng.Sun.COM> cmcmanis@Sun.COM (Chuck McManis) writes:

>I'm quite interested in Biped robots and believe that problem 'b' can be
>solved using a subsumption approach. If you build the system so that the
>foot returns a balance indication (that is, given weight sensors in the
>toes, are all toes seeing the same weight) and tying that to a body
>positioning subsystem whose behaviour is to simply move the center of
>gravity such that the toes are always reading 'in balance', then you've
>basically saved the problem without having to compute any kinematics.

Interesting....

>In such a system, lifting one leg would cause the robot to automatically
>balance on the one 'down' foot. Then push the robot over (add an input 
>to lean forward) and when the other foot hits, it takes over and recovers
>the balance. 

Well, it would most probably fall over unless there was a degree of
"predictive" control between the leg-raising and the balance shift.
The large perturbation of the whole mass having to be supported on
one foot as the other is lifted "unexpectedly" will undoubtedly cause
it to topple.

[Try this: stand on trap-door with one foot, and on floor with the
other.... concentrate ... and let somebody else open the trap door
independently. Now try it with you opening the trap door.  You would
anticipate the loss of support and shift your weight to the foot on the
floor.]

A more fruitful control regime may be to shift the balance to the foot
which is not to be raised (i.e. have it take the weight), then lift the
other foot and allow the balance control system to adjust as the mass
redistributes as part of that action, and the forward motion.

Forward motion is produced by shifting the balance slightly forward,
after the up foot has been raised (a programmed fall), or by the
mechanical action of extending a leg forward.

Then, as the raised foot is placed on the ground (perhaps as it catches
a programmed "fall"), the balance is moved so that it allows the other
foot to be raised, much as before.

A deliberate walking action; extending a leg forward and pulling the
robot forward by then moving the balance above the placed foot seems a
more stable configuration but it would look awkward, being stop-and-go
in nature.

Programmed falling is what most humans do most of the time when
walking.  It's efficient because it maintains momentum in the walk
direction. However, humans do walk by deliberate placement and
rebalancing if the situation demands [ask a burglar :-)].

-- 
Bernd Felsche, MetaPro Systems Pty Ltd
328 Albany Highway, Victoria Park, Western Australia
Phone: +61 9 362 9355  Fax: +61 9 472 3337
