In my last article, "Bipedal Balance;' (Robotics Age, April 1984, p. 21) I described a concept called a Reflex Autonomous Machine. The mechanism mimics the reflexes an animal uses to preserve balance. These reflexes were the vestibular, placing, flexion, and stretch reflexes. They worked by automatically transforming sensor readings into appropriate corrective movements of a robot's legs.
My focus was on bipedal animals such as humans and on bipedal robots. How- ever, these same reflexes serve in quad- rupedal animals. Since I am discussing legged robot "animals,' I could not resist playing the acronym game. My concept for a multilegged robot steering mechanism is called HORNS, a Heuristically Orienting Navigation System. It is designed to reduce the robot's need to use a central control computer to do much more than make very general decisions such as "Keep watching that,' "Investigate that,' or "Carry on.' The HORNS will take care of the steering.
HORNS' underlying principle seems basic to animal steering: the body follows the head. You can see this principle in action very nicely in the praying mantis. When a mantis sees an insect off to one side, its eyes are not stimulated equally. The eye closer to the prey records a larger, brighter image than the other. The head rotates until the eyes' images are equal and the head's axis of symmetry is aimed at the prey.The body then swings around behind the head to bring the forelimb grabbers in to line with the prey. This reflex sequence is controlled by signals from a clump of bristles on each of the mantis' shoulders which are disturbed or bent as the head moves. When head and body are in line, the bristles on the two shoulders are equally disturbed, their signals show no difference, and the forelimbs are triggered to seize the prey. All is automatic. The mantis' puny brain need only say, "Looky there!" and all else follows.
Mammals such as cats, dogs, and humans follow the same principle in another way. They use the tonic neck reflexes to generate changes in limb configuration in response to head and neck movement. As the head bends downward, toward the chest, sensors in the neck respond and the forelimbs flex while the hindlimbs extend. If the head bends upward, the forelimbs extend and the hindlimbs flex. The effect is to line the body up with the head as the head dips to examine some object or to facilitate movement on a grade. If the head turns to one side, the limbs on the side toward which the jaw points (the jaw-side) both extend and the others flex, as shown in Figure 1. These actions are also evidenced when a baseball player goes after a fly ball or a fencer takes the en garde position
The lateral tonic neck reflex is the one most useful in steering. It need not actually produce limb movement, for it is generally conspicuous only under special conditions. But even in humans, it changes the tone of the limb muscles and aids the motor neurons, the final common path, so that willed movements following the pattern of the reflex are easier to perform. In fact, the reflex may be best interpreted as a readiness to move in a particular way.It steers because a reflex-stiffened (jaw-side) limb will tend to function as a pivot around which the opposite limb can step.
Figure 2, which shows a dog getting to its feet and turning, demonstrates this steering effect. As the dog rises and its head begins to turn, the jaw-side limb extends. Then, starting in frame 5, that limb serves as a more or less rigid pivot about which the animal steps. As the dog completes its turn and the head begins to straighten out on the body (in effect moving toward the opposite side) the limb roles reverse. The tonic neck reflex seems to unbalance ordinary linear locomotion in such a way as to make a turn easier to perform in the direction indicated by the head's orientation.
The tonic neck reflex works similarly in humans. As the head turns, the jaw-side leg stiffens while the other steps around it, and the body turns to follow the head. Sometimes, however, the reflex can cause problems. A bicyclist or car driver knows how a sudden turn of the head can cause a swerve-largely because of the reflex effect on the limbs. Humans must sometimes deliberately overrule their reflexes. Fortunately, they can.
Clearly, the lateral tonic neck reflex could be very useful to a robot. Whenever it directs its instruments or eyes toward a. particular object or direction, the body would follow along. To turn left or right, it would only need to look left or right. The turn would follow as the reflex appropriately biased the standard, straight-line locomotor program.
The nonlateral tonic neck reflexes can play other roles. Suppose that a quadrupedal robot is walking toward some object on the ground. As it approaches, it lowers its head to keep the object in view. As the head bends downward, the tonic neck reflex flexes the forelimbs, bringing the eyes nearer to the object, while it stiffens the hindlimbs, stopping movement. If, as the robot approaches, the object flies off into the sky, the head bends upward, the hindlimbs flex, and the forelimbs stiffen. Again the machine stops and sits down, staring doggishly after its elusive goal.
As long as we rule out some external controller, our next interesting question is then, "What makes the head move?" For animals, there are several answers: The head moves, or is prepared to move, in accord with eye movements, gravitational effects on the otoliths (linear accelerometers in the inner ear), and movement effects on the semicircular canals (angular accelerometers in the inner ear).
The otolith effects are simplest. When the head is not level, the pull of gravity on the otolith organs produces signals which bring the head back to that orientation. If the neck is kept from bending, the limbs flex and extend to level the entire body. If the neck is free to bend, the otolith effects then take care of leveling the head on the neck, and the limbs and body then follow the tonic neck reflexes.
You can see this interaction of otolith and tonic neck reflexes in the righting reflexes. If you lay a horse on its side with its head flat on the ground, the otolith response first brings the head upright, the upper ear against the upper shoulder. This evokes a tonic neck reflex that extends the upper limbs and flexes the lower. As the otolith reflex then levels the head to near the normal position, the trunk twists and a second tonic neck reflex extends the lower limbs and brings the animal to its feet. The sequence is coordinated at the lowest level of the brain, and (at least in the horse), is obligatory. The initial head movement must occur if the animal is to get to its feet at all. If you prevent the initial head movement by holding the head down, the animal cannot rise.
The semicircular canals and their responses to angular acceleration are very useful in maintaining balance, but they can also help steer. In this context, we need only consider horizontal acceleration such as an animal would encounter on a turntable or during a turn. The reflex response to such movements is a repetitive oscillatory movement of the eyes, head, and ears called nystagmus. It consists of a fast flick in the direction of rotation, followed by a slow drift in the opposite direction. This reflex serves to prevent apparent movement of the environment by maintaining sensory fixation on environmental features. Nystagmus also occurs in the laboratory when an experimenter rotates the environment around an animal and in nature whan an animal watches a stream of movement such as a brook, drifting clouds, or a line of other animals. Here, though, it is not a reflex response to fluid moving in the semicircular canals. Instead, it is an automatic product of the brain's attempt to fix on a moving stimulus. If the stimulus is visual, the eyes and head will track it; they will also track sounds and odors. The result is a sensorikinetic nystagmus that keeps the head aimed at the stimulus. Tonic neck reflexes ensure that the animal is always ready to pursue the moving stimulus, a readiness of obvious utility to any predator-or to a tennis player.
Still other reflexes may link eye movements to head movements and thence body movements. The eye muscles have stretch receptors much like those in other muscles that serve the stretch reflex. When the eyes move, certain of the eye muscles are stretched, and the stretch receptors generate signals that can affect the neck and limb muscles. The effect of this reflex seems to be to prepare the body to move away from whatever has attracted the gaze. Perhaps it is an avoidance mechanism, reflecting a basic evolutionary axiom; those animals whose first reaction to anything that attracts their attention is fear and readiness to flee may have the best chances of surviving. On the other hand, since a turn of the head must evoke a nystagmic movement of the eyes in the opposite direction, it could be a means of reinforcing the necessary coordination of a turn.
A robot might not use a tonic eye reflex in quite the animal's way,since it certainly has no great need to be ready to flee. Thus, a robot's eye reflex might work just like the lateral tonic neck reflex. This may seem redundant, but movable eyes offer a low mass, quick-response, low-energy way to prime the body for a turn. They also tie steering a bit more directly to visual search, fixation, and decision making. It also ties steering to other distance senses. When you hear a noise in the dark, your eyes promptly move toward the source of the noise. Your eyes can serve as an intermediary between hearing-or other senses -and steering.
All the reflexes we have discussed are imitable. Yet it might be wasteful to try to equip a robot with all of them. After all, a robot is more simply designed than a cat or a human being and its sensors may not move independently. Let us, then, picture a bipedal or quadrupedal robot whose head can turn, but whose distance sensors (for vision and sound) are mounted rigidly on the head. It seems fairly straightforward to design the robot's circuitry so that when it turns its head to fixate on some stimulus, reflexes like the tonic neck reflexes prepare its limbs to move toward the stimulus as soon as the central computer decides to move. An intermediary tonic eye reflex would be necessary only if the sensors could move in or on the head. This reflex might be necessary for large robots equipped with rotating radar scanners or the like
Tying such reflexes into the balance system could provide a very useful way to get a fallen robot back on its feet. As in the horse, there need only be an otolith-like orientation sensor in the head which triggers reflexes to keep the head level. As the head moves, it will in turn trigger tonic neck reflexes that will bring the body to its feet.
The HORNS concept models animal reflexes to create a robot that can steer itself automatically. The model simplifies the problem of robot motor control to its essentials. The ultimate control becomes simply attentiveness, so that any decision to move will move the robot toward whatever holds its attention. It requires only that the robot have suitable sensors for joint or eye movement, plus circuitry that allows signals from these sensors to affect appropriate motor control centers. The result is that the central computer or outside controller has much less controlling to do. In fact, the central computer need do little more than control where the sen- sors point and say, "Go!" Reflexes take care of all the rest.
Can we build a Reflex Autonomous Machine with HORNS? The basic design, worked out by eons of evolution, is a proven success. But is our technology up to duplicating it? We need legged bodies. We need small, cheap microcomputers to control the body parts, perhaps one for each limb. We need sensors for joint movement, acceleration, and gravity. We have all these. What we don't have yet is the programming that will make all these parts work together as smoothly as an animal's parts. But it need not be long before we have a robot with just that programming, one that can be turned loose to explore the moon or Mars or the bottom of the sea with a minimum of guidance. It will steer itself, mostly stay upright, and rise to its feet if it falls. It will truly be an autonomous machine.
