cyberneticzoo.com

Earlier 1976 version sans stereo cameras.

See a few seconds of McGhee's OSU Hexapod in motion in my walking machine compilation video clip.

Stop Press! 20 Oct 2010: Just found fabulous footage of this walker plus others. 50meg download. mp4 runs for 16 mins. see here.

McGhee used Electric drill's to power the legs, similar to his earlier 'Phoney Pony'.

From Basic Robotic Concepts by John M. Holland, 1983.

The OSU Hexapod
At Ohio State University, Robert McGhee and his associates have spent a great deal of effort in perfecting a six-legged insectlike drive system called the "Hexapod" . This system is not the first of its kind to be constructed. That distinction probably belongs to the General Electric "Quadruped Transporter," which was a manually operated four-legged walker developed in the late 1960s. In fact, McGhee himself developed a four-legged walker at the University of Southern California in 1967 [RH- see 'Phoney Pony' link above]. The OSU Hexapod is, however, far more sophisticated in its control than any of these earlier projects. At this time its only rival exists in the Soviet Union, but a similar machine is under construction in Japan.
The Hexapod is being used to develop and test control hardware and software. The system is not intended to be independent, and so it is driven by ac line power through the use of triac controls. The Hexapod is kept tethered and is made to walk short distances over obstacles. One of the forelegs is instrumented with a set of strain gauges and is used for the more complex functions. This is in keeping with the very well proven research axiom of simplifying the problem to the basics.
As with most basic research, the most promising thing about the Hexapod is not the device itself, but the new concepts that are being generated and/or verified during its development. One of the most fundamental concepts that has been studied is "active compliance." This concept arises because position control alone is not sufficient for a walking machine since the sensor systems of the robot will not likely know the exact level of the surface of the ground at the point of contact of each foot. Additionally, the surface firmness may vary greatly under different feet. For these reasons, the control algorithm must include the force being exerted on the surface as part of the feedback loop. This is done by adding a second term to the error feedback signal of the control loop.5 For a rotational joint, the simple position error is

     ERROR = K X (θc — θ)
where,
     θc is the commanded angle, θ is the actual position,
     K is the feedback gain.
     This can also be stated as
     ERROR = K X E_a
where,
     E_a is the angular error.
With active compliance, a torque term is added and the equation becomes:
     ERROR = (K_a X E_a) + (K_t X E_t)

In this case, the angle and torque are both commanded, and the total error is taken to be a combination of the torque error and the angular error. To understand this consider the case of the vertical plane hip joint of a leg as the leg approaches contact with the ground. If the angular error reaches zero and the foot has not yet touched the surface (the ground is lower than the robot expected), the leg will not stop moving. This is because there is still an error signal to drive it. This signal is supplied by the torque term. Since the foot has not touched the surface, this term has an error contribution to the whole error. Conversely, if the leg touched down prematurely, it would not move all the way to the commanded position as the load torque term would go positive after the torque target was passed. This means that the robot trades off torque for position. This same process is used in the velocity and acceleration control loops of the robot. The ratio of the gains of the two terms (K_a and K_t) gives the compliance ratio. McGhee has found that this factor should best be adjusted for the roughness of the ground. It should be noted that compliance is necessary in the horizontal plane of control as well as in the vertical plane.
Other interesting facts have come to light during the development of the Hexapod. According to McGhee, a walking robot should be more efficient on soft surfaces (such as sand and mud) than a rolling machine. This is because rolling machines (treads included) generate a bow-wave effect. This continuous displacement of material all along the path of motion represents a significant power loss. In actual tests, however, walking machines are far less efficient. This is due to several causes, including the use of worm gears in the joints. McGhee has noted that for a walking robot to be efficient, it must recover the kinetic energy from a limb as it slows the limb's motion relative to the body (especially in the unloaded arc of its movement). Worm gears and most common hydraulic controls are not capable of doing this. Additionally, neither of these is very efficient in the first place! Direct-drive motors with back emf braking and power recovery may provide a partial answer to this problem in the future.
It should be noted that the translation between the desired cartesian forces, motions, and positions and their angular counterparts is nontrivial. The term proprioceptive is used to describe the joint control, and the term exterioceptive is used to describe the vector ground reaction forces. McGhee used Jacobian transforms to develop the relationships between these two systems, but the explanation of these is beyond the scope of this discussion.

Leg detail – motors are modified power drills.

From George A. Bekey – Autonomous Robots – From Biological Inspiration to Implementation and Control (2005)

The first autonomous quadruped robot in the United States was constructed in the 1960's at the University of Southern California; it was dubbed the Phony Pony [ McGhee's own documents also have the alternate spelling of "Phoney Pony". Others called it "The Californian Horse".] The four legs of this robot had two joints each (hip and knee). They were identical to one another, and the front and back pair were mounted and controlled in the same way. [This fact made our laboratory the subject of numerous jokes, mostly emphasising that this robot proved that engineers did not know the difference between the front and the back of a horse. Of course, students used somewhat less polite language.] This was a highly non-biological structure, since the architecture of the front and back legs of living quadrupeds and their support  structure differe markedly from those of the robot. Nevertheless, the robot was capable of emulating a number of quadruped gasit patterns, including crawl, walk, and trot, but at a very slow speed. It was able to maintain static vertical stability in these gaits by means of unique spring-restrained "pelvic" structure and by virtue of its wide feet. The machine was not capable of high-speed gaits (such as cantor or gallop) in which all four legs may be off the ground for short time intervals.

This robot was constructed before the existance of microprocessors, so it was controlled by means of a remote minicomputer (located on the second floor of an adjacent building!). The cable connecting the robot with the computer is visible in the right edge of the figure. The robot was controlled by a finite-state machine using sensory feedback on the state of its joints, without any internal model of its kinematics or dynamics.

Al;though the Phony Pony was constructed by Andrew Frank (1968) as part of his Ph.D. dissertation, [I {Bekey} was chairman of Andrew Frank's doctoral committee, but his work was directed largely by Bob McGhee and Rajko Tomovic. McGhee and I obtained a visiting-scholar grant from the National Science Foundation that enabled Tomovic to spend a year at USC, Frank then spent a postdoctoral year with tomovic in Yugoslavia.] much of the theoretical work was done by Bob McGhee and Rajko Tomovic, who proposed a finite-state machine as a model for human locomotion. This work was then elaborated as a control method for quadruped locomotion and used by Frank for control of the Phony Pony. The theory was further elaborated with respect to quadruped "creeping gaits", that is, gaits that can be executed while keeping at least three feet on the ground at all times. (Although in theory there are many possible quadruped gaits, only six can be considered creeping gaits.) Frank and McGhee then established under what conditions a trot can be made stable using a control system whose only inputs are leg joint angles.

The control of the Phony Pony probably represents the first application of finite-state automata to robot walking, a number of years before the development of Genghis at   MIT.