BIOLOGICAL INSPIRATION FOR COMPUTING 273
and controlled travel.^69 With this basic unit, a two-legged running animal (the Planar Biped) could be
modeled as a body with two pogo sticks working 180° out of phase.^70 A four-legged animal could
consist of two two-legged pairs working in opposition (left front and right rear, for example).^71
Since Raibert’s pioneering work, these insights have been applicable to the design of other artificial
legged locomotion devices. For example, an autonomous hexapod named “RHex” has a motor associ-
ated with each leg, each of which is springy and is able to turn on its central axis. This design enables
RHex to have self-correcting reflexes that enable it to respond to obstacles without computational
control. Another family of six-legged robots, called the SPRAWL family, is cockroaches. Each leg,
driven by a piston, acts as a spring that enables SPRAWL robots to bounce over objects in their path
without feedback from the environment. Analysis of the force pattern exerted by the legs closely matches
that exerted by a running cockroach.
Other robots are intended to manipulate objects into precise orientations. The traditional way to
build such robots is to build them rigidly, with limb motion effected through motors and gear assem-
blies to increase torque. However, gear assemblies are inherently imprecise, because their very motion
requires some degree of play where the gears meet (i.e., some nonzero compliance). In practice, the
effect of compliance in the gears introduces a noise function that greatly complicates the prediction of
how a limb will move given a certain motor input, and puts limits on the precision with which the final
orientation can be known.
One solution to this problem is to use “direct-drive” motors placed at every joint, thus eliminating
the gears entirely.^72 Another solution is based on the deliberate introduction of compliance into a gear
assembly. This solution is based on the observation that humans can effect precise positioning without
precision in their joints. In particular, natural joints are often based on ball-and-socket mechanisms even
when they are intended to exhibit 1 degree of freedom. Soft tissue around and in the ball joint intro-
duces compressive compliance in the joint, allowing it to absorb impact and automatically maintain a
degree of tightness in the joint.
In the robot context, Pratt et al. inserted a spring mechanism into a limb joint so that the response
lags the input.^73 This spring adds a large but known compliance in series into the joint (so-called series
elasticity) that is much larger than the unknown compliance of the gears; thus, the gear compliance can
safely be ignored in the prediction of final position. Entirely apart from the increased ease of prediction,
the introduction of series elasticity enables a local response to any sudden changes in loading—during
which time the motors involved can build up torque to handle that load. Other benefits include shock
tolerance, lower reflected inertia, more accurate and stable force control, less damage during inadvert-
ent contact, and energy storage.
8.3.3 Neuroscience and Computing,
Natural brains demonstrate an alternative to the traditional von Neumann computing architecture
(i.e., a fully serial information processor); thus, it is natural to consider possible lessons of neuroscience
for computer design. These lessons occur at varying levels of detail.
(^69) See http://www.ai.mit.edu/projects/leglab/robots/2D_hopper/2D_hopper.html; see also M.H. Raibert and H.B. Brown,
Jr., “Experiments in Balance with a 2D One-legged Hopping Machine,” ASME Journal of Dynamic Systems, Measurement, and
Control 106:75-81, 1984.
(^70) See http://www.ai.mit.edu/projects/leglab/robots/2D_biped/2D_biped.html; see also J. Hodgins and M.H. Raibert, “Pla-
nar Biped Goes Head Over Heels,” Proceedings ASME Winter Annual Meeting, Boston, December 1987.
(^71) See http://www.ai.mit.edu/projects/leglab/robots/quadruped/quadruped.html; see also M.H. Raibert, “Four-legged Run-
ning with One-legged Algorithms,” pp. 311-315 in Second International Symposium on Robotics Research, H. Hanafusa and H. Inoue,
eds., MIT Press, Cambridge, MA, 1985.
(^72) H. Asada and T. Kanade, “Design of a Direct-Drive Mechanical Arm,” ASME Journal of Vibration, Stress, and Reliability in
Design 105(3):312-316, 1983.
(^73) G.A. Pratt, M.M. Williamson, P. Dillworth, J. Pratt, K. Ulland, and A. Wright, “Stiffness Isn’t Everything,” preprints of the
Fourth International Symposium on Experimental Robotics, ISER ’95, Stanford, CA, June 30-July 2, 1995.