Lucy MkI used simple servomotors rigged up to behave roughly like pairs of muscles. However, these had many problems:

• They were too weak to lift a useful payload
• They were very noisy (making it difficult for Lucy to learn to hear and speak)
• They reacted too slowly, making them stiff when they were supposed to be relaxed, and twitchy under tension.

For Lucy MkII we needed far more powerful and responsive muscles, with more biologically realistic qualities.

This latter point is important. The vast majority of robots use 'stiff' motor systems - their limbs are directly connected to the gearboxes of the motors. This is very unlike the muscles of animals in several ways:

• Animals can let their limbs go loose and floppy. This allows them to swing in a natural rhythm, and in humans allows a mother, say, to guide her child's arms and demonstrate everything from how to draw a triangle to the right way to swing a nine iron.

• Animals can feel resistance and external pressure on their limbs by measuring the stretch in their muscles. This is an important source of information for building a model of the animal's position in space. Many robots will just jam solid if an arm catches on an obstruction - the robot simply doesn't know it is there.

• Most robots move in very 'mechanical' ways, quite unlike the fluid ballistic movements of animals. If we are to understand how the brain controls movement it is very important that Lucy's body emulates a biological body as much as possible. Almost certainly the way the brain controls movement relies strongly on the particular qualities of muscles and tendons. Moving such complex and non-linear mechanisms is very much harder in the first instance, but far more likely to provide useful insights than trying to bypass the problem by building high-precision, highly predictable motor systems.

Unfortunately, there are no current technologies that can match animal muscle for length of stroke, speed, elasticity, strength and efficiency. There are many materials under development - memory alloys, hydrogels, pneumatics, etc. - that may fit the bill one day, but for the moment ordinary DC motors are about the only easily accessible source of motive power. The problem is how to make motors emulate muscles.

My first solution was to use an elastic actuator, similar in concept to those used in MIT's Cog robot. This had both practical and theoretical problems.

The second solution was really neat - much simpler, cheaper and more biological in its qualities. However, so far I've been unable to reduce the vibration to a satisfactory level, given the engineering facilities available and the space constraints in the robot.

Sadly I couldn't perfect this method and eventually ran out of time and money, so finally I've plumped for a third solution which is nowhere near as nice but has the considerable advantage of actually working!

Copyright © 2004 Cyberlife Research Ltd.
Last modified: 06/04/04