Figure 1.
Humans have built aircraft, submarines, and other machines that imitate or improve upon animal locomotion, but the design and construction of feasible walking and running machines remains a challenge. From a practical perspective, legged robots offer several potential advantages over wheeled vehicles. They can traverse rough terrain by stepping across or jumping over obstacles using isolated ground contacts rather than the continuous path of support required by wheels. This agility is important in difficult environments with few suitable footholds. Unlike wheeled vehicles, legged robots have an active suspension system that can decouple variations in the terrain from the motion of the body and provide a steady platform for a sensor or payload. From a scientific perspective, researchers would like to construct legged vehicles with the capabilities of animals to better understand the principles of locomotion in biological systems. Over the past hundred years, researchers have combined innovative engineering with scientific observations of the agility and efficiency of legged animals to construct many types of legged mechanisms.
Walking and running machines are divided into two categories based on the stability of their motion: passively stable and dynamically stable. The vertical projection of the center of gravity of a passively stable system always remains within the convex region formed by the contact points of the feet on the ground. This region is called the support polygon. Statically stable machines can stop moving at any time in the locomotion cycle and maintain balance. They typically have four or six legs but may be bipeds with large feet. In contrast, dynamically stable systems utilize dynamic forces and feedback to maintain control and are stable in a limit cycle that repeats once each stride. Dynamically stable machines have been built with one, two, and four legs. Because dynamically stable systems are more difficult to design and analyze, the early development of legged robots focused on statically stable machines.
The earliest walking machines used gears to produce fixed patterns of leg motion for walking. The fixed patterns prevented these machines from responding in a flexible fashion to variations in terrain. Nevertheless, their construction initiated the study of leg mechanisms and gait patterns and they remain useful comparisons for the evaluation of current robots. A more agile walking machine was built in 1968 by Ralph Mosher at General Electric (Liston and Mosher 1968). An eleven-foot-tall, three-thousand-pound machine with twelve degrees of freedom, the walking truck was hydraulically powered and capable of climbing over obstacles, pushing large objects, and walking steadily at five miles per hour. The machine was controlled by a human driver who used his arms to control the front legs and his legs to control the rear legs of the machine. Force feedback allowed the driver to sense the balance of the system, but the machine was difficult and tiring to control even after substantial training. Because the machine was human controlled, it was not an autonomous legged vehicle, but it provided a convincing demonstration of the agility possible with a mechanical legged system.
Figure 1.
Digital computers provided the numerical computation necessary to develop automatically controlled walking robots in the 1970s. Computers were used to determine footholds that maintained stability, to solve the kinematic equations for positioning the legs, to provide feedback control based on the body orientation and leg positions, and to plan paths through the environment (see PLANNING). Several of these early machines were hexapods because six-leg designs allow walking with a stable, alternating tripod gait. The Ohio State University Hexapod was built in 1977 by Robert McGhee using electric drill motors (Bucket 1977; McGhee 1983). A second hexapod from OSU, the Adaptive Suspension Vehicle, was a three-ton vehicle with a human driver but automatic positioning of the legs under most terrain conditions (Song and Waldron 1989) as shown in figure 1. The first computer-controlled, self-contained walking robot was a hexapod, built by Sutherland in 1983 (Raibert and Sutherland 1983). The first commercially available legged robot was the ODEX 1 "Functionoid" built by Odetics in 1983 (Russell 1983).
Research in legged locomotion has proceeded primarily along two lines: leg design and control. Design considers the geometries of leg arrangements and mechanisms to increase motion, strength, speed, or reliability, decrease energy requirements or weight, or simplify control. For example, Shigeo Hirose's 1980 PV II quadruped used a pantograph leg mechanism that allowed actuators independently to control each degree of freedom in Cartesian space, considerably simplifying the kinematic equations (Hirose and Umetani 1980). The Ambler, a thirty-foot-tall, hexapod planetary explorer built at Carnegie Mellon University, used an orthogonal leg design to simplify the kinematic equations (Simmons et al. 1992). The design kept the outer segments of the legs vertical while positioning them with inner segments that rotated in a horizontal plane. Dante, also built at CMU, used sets of legs mounted on sliding frames. After one frame's legs lifted, the frame slid forward and then lowered the legs for the next step. In 1994, Dante II successfully descended into an active volcano in Alaska and analyzed high temperature gases (Wettergreen, Pangels, and Bares 1995).
Issues of control have been explored most often in bipedal walking and running robots. Computer-controlled bipeds that used large feet to allow static stability were built in 1972 (Kato and Tsuiki 1972). These were followed by quasi-static walkers that included a dynamic phase during which the machine fell forward onto the next stance foot (Kato et al. 1983). The first dynamically stable walking bipeds were designed by Miura and Shimoyama (1984). These machines used stiff legs that were raised by rotating the hip and produced motion that resembled walking on stilts.
Figure 2. The planar biped running right to left, up and down a short flight of stairs. Photograph reprinted by permission of the MIT Leg Laboratory.
In the early and mid-1980s, Raibert and colleagues at MIT and CMU designed dynamically stable running monopods, bipeds, and quadrupeds. These machines used springs in the legs to provide a passive rebound during the stance phase and hydraulic actuators to provide thrust and to control the leg angle (Raibert 1986). The control systems for these machines divided the complex system dynamics into three largely decoupled problems: hopping height, forward speed, and body attitude. Hopping height was maintained by extending the actuator in series with the leg spring. Body attitude was controlled by applying a torque at the hip while the foot was in contact with the ground. Forward running speed was controlled by positioning the foot at touchdown. A finite state machine determined the active control laws at any given moment. In addition to running, these machines also used a variety of gaits (Raibert, Chepponis, and Brown 1986), ran fast (Koechling and Raibert 1988), performed flips (Hodgins and Raibert 1990), and ran up and down stairs (Hodgins and Raibert 1991), as shown in figure 2.
Further exploration of control for legged robots has led to the discovery of an elegant class of legged robots with no computer control at all. Called passive walkers, these machines walk down an inclined slope but have no internal source of energy. The length of the legs and mass distribution are chosen so that the legs swing forward and the knees straighten for touchdown without actuation (McGeer 1990a, 1990b). Controllers for passive running have also been proposed (McGeer 1989; Ringrose 1997; Ahmadi and Buehler 1995). Recently, several researchers have begun to build a formal theoretical framework for analyzing the stability of dynamic running machines. This approach models dynamic robots as simpler systems such as spring/mass systems and then proves stability of the model given a particular set of control laws (Koditschek and Buehler 1991; Schwind and Koditschek 1997).
Although the dream of an artificial legged creature such as C-3PO from the movie Star Wars has not been realized, research into running and walking machines has furthered our understanding of legged locomotion in machines and animals. Researchers have explored a variety of geometries, mechanisms, control techniques, gaits, and motion styles for legged machines, and the resulting insights have enabled new applications and designs, as well as a growing theoretical foundation for legged locomotion and control.
The planar biped running right to left, up and down a short flight of stairs. Photograph reprinted by permission of the MIT Leg Laboratory.
Ahmadi, M., and M. Buehler. (1995). A control strategy for stable passive running. Proceedings of the International Conference on Intelligent Robots and Systems. Los Alamitos, CA: IEEE Computer Society Press, pp. 152-157.
Bucket, J. R. (1977). Design of an On-board Electronic Joint System for a Hexapod Vehicle. Master's thesis, Ohio State University.
Hirose, S., and Y. Umetani. (1980). The basic motion regulation system for a quadruped walking machine. American Society of Mechanical Engineers Paper 80-DET-34.
Hodgins, J., and M. H. Raibert. (1990). Biped gymnastics. International Journal of Robotics Research 9:115-132.
Hodgins, J., and M. H. Raibert. (1991). Adjusting step length for rough terrain locomotion. IEEE Transactions on Robotics and Automation 7:289-298.
Kato, T., A. Takanishi, H. Jishikawa, and I. Kato. (1983). The realization of quasi-dynamic walking by the biped walking machine. In A. Morecki, G. Bianchi, and K. Kedzior, Eds. Fourth Symposium on Theory and Practice of Robots and Manipulators. Warsaw: Polish Scientific Publishers, pp. 341-351.
Kato, I., and H. Tsuiki. (1972). The hydraulically powered biped walking machine with a high carrying capacity. In Fourth Symposium on External Extremities. Dubrovnik: Yugoslav Committee for Electronics and Automation.
Koditschek, D. E., and M. Buehler. (1991). Analysis of a simplified hopping robot. International Journal of Robotics Research 10: 6.
Koechling, J., and M. Raibert. (1988). How fast can a legged robot run? In K. Youcef-Toumi and H. Kazerooni, Eds., Symposium in Robotics, DSC, vol. 1. New York: American Society of Mechanical Engineers.
Liston, R. A., and R. S. Mosher. (1968). A versatile walking truck. Proceedings of the Transportation Engineering Conference. London: Institution of Civil Engineers.
Miura, H., and I. Shimoyama. (1984). Dynamic walk of a biped. International Journal of Robotics Research 3.2:60-74.
McGeer, T. (1989). Passive bipedal running. Centre for Systems Science, Simon Fraser University, Technical Report CSS-IS TR 89-02.
McGeer T. (1990a). Passive dynamic walking. The International Journal of Robotics Research 9:62-82.
McGeer, T. (1990b). Passive walking with knees. Proceedings of the 1990 IEEE Robotics and Automation Conference, Cincinnati, OH, pp. 1640-1645.
McGhee, R. B. (1983). Vehicular legged locomotion. In G. N. Saridis, Ed., Advances in Automation and Robotics. JAI Press.
Raibert, M. H. (1986). Legged Robots That Balance. Cambridge, MA: MIT Press.
Raibert, M. H., M. Chepponis, and H. B. Brown, Jr. (1986). Running on four legs as though they were one. IEEE J. Robotics and Automation 2:70-82.
Raibert, M. H., and I. E. Sutherland. (1983). Machines that walk. Scientific American 248(2):44-53.
Ringrose, R. (1997). Self-stabilizing running. Proceedings of the International Conference on Robotics and Automation, Albuquerque, NM.
Russell, M. (1983). Odex 1: The first functionoid. Robotics Age 5(5):12-18.
Schwind, W. J., and D. E. Koditschek. (1997). Characterization of monoped equilibrium gaits. Proceedings of the 1997 International Conference on Robotics and Automation, Albuquerque, NM, pp. 1986-1992.
Simmons, R., E. Krotkov, W. Whittaker, and B. Albrecht. (1992). Progress towards robotic exploration of extreme terrain. Applied Intelligence: The International Journal of Artificial Intelligence, Neural Networks and Complex Problem-Solving Technologies 2(2):162-180.
Song, S. M., and K. J. Waldron. (1989). Machines That Walk. Cambridge, MA: MIT Press.
Wettergreen, D., H. Pangels, and J. Bares. (1995). Behavior-based gait execution for the Dante II walking robot. Proceedings of the 1995 IEEE/RSJ International Conference on Intelligent Robots and Systems. Los Alamitos, CA: IEEE Computer Society Press, 274-279 .
Copyright © 1999 Massachusetts Institute of Technology