Proc. IEEE Int’l. Conf.

on Robotics and Automation, New Orleans, LA, April 26-May 1, 2004

An Actuator with Physically Variable Stiffness for Highly Dynamic Legged Locomotion
Jonathan W. Hurst, Joel E. Chestnutt and Alfred A. Rizzi
Carnegie Mellon University Email: {jhurst, chestnutt, arizzi+}@ri.cmu.edu

Abstract— Running is a complex dynamical task which places strict design requirements on both the physical components and software control systems of a robot. This paper explores some of those requirements and illustrates how a variable compliance actuation system can satisfy them. We present the design, analysis, simulation, and benchtop experimental validation of such an actuator system. We demonstrate, through simulation, the application of our prototype actuator to the problem of biped running.

I. I NTRODUCTION A robot designed specifically for autonomous legged locomotion should be capable of highly dynamic running, jumping, and stumble recovery, and should achieve these capabilities while being energetically efficient. In order to accomplish these goals, we believe a leg spring of sufficient capacity to store the energy of a running gait is a necessity. Furthermore, we believe variable leg spring stiffness provides an important means for effective gait control. One could concieve of two extremes of actuator design that would create these desirable properties: high-bandwidth actuators with all dynamics described by software control policies, or carefully designed mechanical systems with tuned natural dynamics that require no software control. The first method is flexible, although actuators with sufficiently high bandwidth and power capacity may not exist for the locomotion task. The second approach is quite inflexible, and requires extensive knowledge of the desired behavior before construction, but there is no bandwidth limit preventing the mechanism from behaving as intended at high frequencies. This method is taskspecific, and the resulting designs are unlikely to be able to perform the breadth of tasks required for running. Our actuator design, shown in Figures 1 and 4, represents a carefully chosen balance between the two design extremes. The natural dynamics of the system are carefully designed and are utilized where possible, while the software controller adds energy that is lost in the mechanism, and creates behaviors that are not inherent in the natural dynamics. The actuator exhibits natural dynamics that are similar to those of animals, and is based on a previously developed mathematical model of running, shown in Figure 2, the spring loaded inverted pendulum (SLIP) [1], [2], [3]. II. BACKGROUND Most research papers that analyse the mechanics of running base this analysis on some form of the SLIP model. This model describes the motion of the center of mass of a running animal [4], [5] or a running robot. The basic definition of running [6]
Fig. 1. The actuator prototype, mounted rigidly at the hip to an optical table

M X K q

Fig. 2.

The Spring Loaded Inverted Pendulum model of legged locomotion

is linked to the SLIP idea – energy is transferred from kinetic energy in the flight phase to spring energy in the stance phase, and vice versa. The natural physical instantiation of the SLIP model utilizes passive leg springs for this energy storage. A. Compliance and Running Physical series compliance is virtually necessary to achieve a successful running gait. Simulating compliance using a rigid actuator such as an electric gearmotor is not feasible for three reasons: bandwidth limitations, power output limitations, and energetic efficiency. The bandwidth limitation of an electric motor is due, in large part, to the high reflected inertia linked rigidly to the robot leg, making a correct dynamic response to impacts impossible. The power density of a physical spring is arbitrarily high, depending on its stiffness, making a compelling argument for combining the relatively high work capacity of a motor and power source with the high power density of a physical spring. Springs are particularly useful in rhythmic systems, because energy can be stored and released

Throughout the design process. One motor controls the spring pretension. spring position as a function of z deflection of the spring. r2 θ2 − r1 θ1 deflection of the cable after the pulley function: (x3 + ∆x) or (x3 − ∆x) knee force. The mass of this actuator prototype is approximately 4kg. τeff /r2 spiral pulleys. [14]. hopping height. These two parameters. Another choice might be leg stiffness. which are very light for a given load rating. in two different forms: one rotational. because the desired spring function was unknown. batteries. Throughout the remainder of this paper we will use the notational conventions of Table 1. and because two logarithmic spirals mesh correctly and provide a stiffening function [20]. and leg stiffness directly affects one or more of these parameters. ∆x) G(z ) y Fy (y ) Fz (z ) (b) Linear model Mechanical models of the physical actuator description Motor position leg position motor inertia leg inertia pretension linear leg deflection. For mathematical analysis in [1]. however. 3. Keff . all joints contain thin-section bearings. Animals store mechanical spring energy during a running gait. ground stiffness. variable θ1 θ2 J1 J2 x3 ∆x z Feff (x3 . touchdown angle. corresponding to spring stiffness (x3 in Figure 3(a)). III. [12]. There are two degrees of freedom. hopping or running on a surface of changing stiffness [4]. [11]. All parts are machined aluminum. therefore the mechanical design is an essential part of the overall control system design. and flight duration was demonstrated experimentally in [19]. before the pulleys force function of the spring force on the cable after the pulley function TABLE I position (θ1 in Figure 3(a) and x1 in Figure 3(b)). Control of forward speed. Energy is stored in fiberglass plates linked to spiral pulleys. It is influenced by changes in leg stiffness (caused in animals by muscle co-contraction and limb geometry) [12]. A. and new pulleys can be designed to provide an arbitrary spring function. to create a nonlinear spring function. T HE ACTUATOR D ESIGN The natural dynamics of a system are an inseparable part of its behavior. [9]. Physically Variable Compliance as a Method of Control While physical compliance is virtually a necessity for successful running. hopping in place with varying frequency [4]. using other methods to change global vertical stiffness. The actuator is essentially a single compliant joint. [13]. Given . The actuator is intended as a prototype leg for a bipedal robot with approximately 1 meter leg length and 30kg total mass. or running at different speeds with constant stride length [17]. 20kg are reserved for motors. x3 . [8]. [18]. This model is illustrated in Figure 3. and stride length. most likely for these reasons and more [7].much more efficiently through a spring than if it were passed through the motor. the dynamics of the system controlling the pretension. [16]. In both models. and computing. Animals have physically variable leg compliance (tuneable natural dynamics). are the parameters necessary to control SLIP model running. they do change leg stiffness when other methods are not available. are ignored and in the linear model. Of this 30kg. [13]. stance duration. and power electronics with each transfer. [15]. and it is nearly 50% oversized. The reduction ratio of the pulleys varies proportionally with the fiberglass spring deflection. there are many possible parameterizations that could be considered. and vary it to control running and hopping in certain situations [3]. [17]. [13]. The rotational model is physically similar to the prototype actuator. The global vertical stiffness is used by animals to control running speed. [4]. [11]. and two corresponding motors. [10]. transmission. To minimize weight. For example. with mechanical design choices made to match the mechanical model upon which the software controller is based. while the linear model is a simpler form that still captures the important properties of the system. along with the leg angle at touchdown. The running gait of a SLIP can be described by three parameters [1]. while the other motor controls the spring rest tmotor J1 r1 x3 B1 q1 B1 Fmotor X1 M1 Fz(zB) J2 B2 r2 Fz(z)A Keff q2 X2 M2 B2 Fleg tleg (a) Rotational model Fig. and the number of legs on the ground at once. global vertical stiffness of the center of mass is the important factor [3]. minimizing weight has been a priority. and increases with the square of the running speed [16]. The actuator presented here was carefully designed as an integrated system of mechanism and software controller. In a running gait. endowed with tuneable natural dynamics. Mechanical Design Perhaps the most important aspect of the actuator is the physically variable series compliance. Most research suggests that animals prefer to maintain leg stiffness over a range of running speeds [3]. These pulleys are modular components. is assumed to simply be a programmable value. varying the compliance provides a useful tool for gait control. Logarithmic spiral pulleys were chosen because the spring function of the bending fiberglass plates was unknown. the spring stiffness. including springs and power transmission. B. This allotment leaves only 10kg for the entire framework and mechanism. one linear. leg length and angular velocity at bottom of stance along with the leg stiffness were chosen as the three parameters. with the main structural members (analogous to the femur) made of thinwall aluminum tube.

They are not constrained to a single degree of freedom. allowing the motors to be located remotely. Adjusting spring rest length is accomplished using a simple PD controller on the position of x1 . zero-backlash system. The remaining four pulleys are floating. but diagrammatically located near the motor. FzA B2 J 2 G(zA) q2 r2 t leg G(zB) yB Fy(yB) Fy(yA) zB. Refer to Table 1 for notation descriptions. There is a speed reduction between θ1 and θ2 . θ1 . the spiral pulleys are also pinned in place but free to rotate. 5. when the mechanism is either out of adjustment range or in transition to the correct configuration. placing two such spring/pulley systems in direct opposition results in a single effective torsional spring whose stiffness is determined by the pretension on each individual nonlinear spring. tasks. we can solve this equation for x1 to calculate the desired position. and knee joint damping.a general spring force function Fy (y ) and a general pulley transmission function y = G(z ). In the ideal case. no further control is currently required. All friction related to the speed reduction is applied to θ1 and corresponds to B1 . minimizing friction. The ideal case of our actuator. and can move sideways as well as rotate. In order to create a low-friction. Cable routing diagram of the actuator. 5. To simulate this system. Fz = Kz 2 ). shown in Fig. x1 is fixed at the desired set point x∗ 2 . and the spring physically matches the desired spring stiffness K ∗ . while the inertia corresponds to J1 . FzB x3 Fig. for example. Adjusting the spring stiffness is accomplished with a simple PID position controller on the pretension motor. such as during the flight phase of a running gait. Because the specific position of this motor corresponds to a specific effective stiffness Keff . so it has been added to the model. The second is to actively control the motor (x1 ) to simulate the proper settings. we must control x1 so that it will simulate the desired spring stiffness if the physical system does not match our desired system. though it can easily be set to zero. The cables may easily be routed around joints and may flex in two planes. In the ideal setting. The speed reduction is physically located near the knee joint. like standard belts or gears. where x3 represents the pretension on the two nonlinear springs and ∆x represents the deflection from their rest position. 4. The torque on the leg applied by the physical system should match the torque on the leg applied by the desired system. it is implemented using a combination of a block-and-tackle pulley mechanism and a difference in diameters between θ1 and θ2 . ∂z allowing for suitable shaping of the overall effective spring function through choice of the pulley transmission function G(z ). A speed reducer also amplifies the motor inertia by the square of the speed reduction. There is some damping in the real system. the force as a function of z is ∂G (z ). Keff (∆x. x3 )r2 + . x∗ 2 . and will allow the mechanical springs to store and release most of the energy in a running gait. our actuator utilizes a high-speed cable drive [21]. We ignore load forces and use a much larger motor than necessary for the current prototype. spring stiffness. backlash. K* M2 X2* B2* 2 r2 X2 Fig. illustrating the role of each motor in the tension of the two springs. After creating a nonlinear spring function (for example. Many of the mechanism design challenges are common ones. low friction and inertia are most important in this part of the actuator. with some added complexity when the pretension x3 is not properly adjusted. Figure 4 shows the cable routing. The low-frequency behaviors of the system are handled by the motor. the motor will have to do very little work. the rate at which this parameter can be varied depends on the actuator and transmission used. J1 and J2 are pinned in place but can rotate freely. not shown on the diagram. and is easily incorporated in the model. the resulting effective spring force is Fz (z ) = Fy (G(z )) Feff = 4Kx3 ∆x. For the example quadratic spring. this amplification appears in the relatively large values of M1 . ∗ B2 B2 ∗ x ˙ 2 = (x2 − x∗ x ˙ 2. Control System Design The control system is designed for the mechanical model shown in figure 3. where z is the extension of the cable out of the pulley. Any stretch in the cable effectively adds series compliance to the system. Also shown is the fact that a displacement of the leg (θ2 or x2 ) results in either displacement of the motor (θ1 or x1 ) or displacement of the springs. Because the high-frequency behavior of the system is generally handled by the springs. Our prototype is intended for relatively slow changes at low force. the stiffness of the resulting system can be changed by adjusting the pretension. B. with three parameters: spring rest ∗ ∗ length. The first is to adjust the mechanism configuration so its physical properties match the commanded spring stiffness and rest length. and no speed reduction. The transmission between θ2 and the springs has very low friction. K . x∗ 1. and thus friction and inertia can be overcome by relatively low-bandwidth software compensation. and inertia. B . 2 )K r2 + r2 r2 Assuming the dependence of Keff on ∆x is linear (recall that ∆x = x2 − x1 ). In this manner. x3 . In practice. and it is possible that some damping will be desired in the ideal system. and is intended to accomplish two basic q1 J1 tmotor r1 B1 yA zA.

along with a spring cancellation force to hold it against the force applied by the springs: Fcom = KP (x∗ ˙∗ ˙ 1 ) − Keff (∆x.3 0. The function was measured using a load cell mounted against the leg of the actuator. x3 18 16 Average Hysteresis (N) 14 12 10 8 6 4 2 0 0. x3 ). by using tests that would be difficult to apply in the real world. instead of the desired behavior. When trying to simulate a stiffness at high frequency. As is shown on the stiffness function graph in Figure 6. or when the mechanism is in the process of adjusting to the correct stiffness. This figure was created by averaging the hysteresis at each of a range of pretension settings. IV. 7.06 0. recording applied force of the leg while cycling through a range of set points at different pretension settings. The second source of error comes from the bandwidth limitation on x1 . resulting in the correct desired output function. because the hysteresis was relatively constant over a range of deflections at a given pretension value. and is not part of a running robot that exhibits all of the desired dynamic properties. ∆x.22 0.x∗ 1 = K∗ ∗ B ∗ − B2 (x2 − x2 ) − 2 ˙ 2 + x2 . We begin by measuring and testing certain aspects of the physical prototype. without the attached spring and associated dynamics. For each pretension value. so these measurements may be used in a software simulation. The level of hysteresis as a function of pretension is depicted in Figure 7.26 0. so the spring hysteresis would account for most of the energy lost to friction in a running gait. The actuator is then compared to the simulation at a range of frequencies. A.04 We then apply a PD controller on x1 to move it to the desired position. 6. which can exhibit energy loss due to seal friction on the order of 25%.2 0 Spring Deflection (m) 0. As can be seen in Figure 6. which violates our earlier assumption. 1 − x 1 ) + KD ( x 1 −x With the spring cancellation force. We used an approximation of Keff that is linear in ∆x. hysteresis increases as the pretension (and forces applied to all bearings and cables) gets larger. and pretension. This error can be remedied by manufacturing new spiral pulleys to create a new Fig. In comparison to air cylinders used in some hopping robots [22]. As in the earlier example.28 Fig.24 Pretension (m) 0. the actuator simulation is tested in a simulation of a bipedal runner. S IMULATION . However. providing evidence that the actuator could be effectively used on a physical running robot. there is some hysteresis due to friction. Further analysis is done using the simulation. Fz (z ). R ESULTS AND C OMPARISON Because this actuator is currently bolted to an optical table. hysteresis was nearly 10% of the largest applied force. The friction is caused by the sliding of the spring guides and the rolling of the bearings on the nonlinear pulleys. There are two limitations in this approach that introduce error. and could be depicted with an additional damper and stiction model in parallel with the springs in Figure 3. this loss is relatively small. As might be expected. 2 x Keff Keff r2 Force (N) 150 100 50 0 0. the PD control can adjust x1 as if it were an independent mass. forces on the order of 180N were applied—so at very high pretension. and the system will revert to the behavior of its natural dynamics. it is necessarily an approximation. so our match of Keff to K ∗ will only be as good as our approximation fits the actual relationship defining Keff . this error can only happen when the desired stiffness is outside the range of the mechanism.02 Pretension (m) 0. x3 G(z ). the calculated location x∗ 1 will only be correct to the accuracy of the approximation. Finally. the inertia of M1 will limit the acceleration of x1 . In addition. Measured spring force as a function of deflection.2 0. . Measured average hysteresis as a function of pretension. because the stiffness of the actuator is adjustable. Static Compliance Characterization We modeled the spring function of the actuator in order to create an accurate simulation.25 0. some experimental evidence is required to support the assertion that it does indeed match the SLIP model and would work effectively in a dynamic legged robot. The slope is not quite linear. First. The knee friction B2 is very small. showing that the software-electricalmechanical system behaves very similarly to the simulation we have developed. because Keff is a function composed of the logarithmic spiral pulleys and the unknown fiberglass spring function. creating a quadratic spring function would result in an overall effective spring constant Keff which is linear and stiffens with pretension changes. when simulating a spring stiffness that is outside the physical range of the actuator. the spring function Feff = Keff (∆x) gets stiffer as the pretension increases.

At low frequencies. the motor is capable of simulating the commanded spring constant K ∗ by compensating actively for the discrepancy. the published properties of our motor.5 0 0.5 Fig. triangles are 5000 N/m. and should be exhibited accurately to a high bandwidth. with set point and stiffness as programmable variables that can change instantaneously. The basic structure of the graphs between the simulator and the actuator is very similar. as output.5 θ1 (data) θ (simulation) 1 Frequency plots of physical and simulated actuator Magnitude (dB) 20 10 0 −10 0 10 Frequency (Hz) θ (data) 2 θ (simulation) 2 K =3000 N/m eff K =4000 N/m eff Keff=5000 N/m 1 1. Magnitude and phase of the ratio of actual to desired impedance as a function of frequency. x1 . which converges to the natural dynamics of the system. The motor position.5 Leg Position (rad) 0. one lower.0517 N·m·s/rad 0. Figure 10 shows the ratio of expected to desired impedance of the actuator 1 SolidWorks simulation measured over a range of frequencies. Circles are 3000 N/m. and the same as the physical system stiffness Keff TABLE II P HYSICAL PROPERTIES USED FOR SIMULATION To illustrate performance over a range of frequencies. the motor does virtually nothing to affect the exhibited spring stiffness. it does serve to compare it and the simulation and verify that both behave similarly at a range of frequencies. Data points are depicted on experimental data from the physical actuator. At very high frequencies. Keff . One major benefit of this actuator design is that aspects of its natural dynamics are programmable. x1 . with the commanded spring stiffness K ∗ higher. stars are 4000 N/m. variable J1 J2 B1 B2 value 0.5 kg 0. along with the measured stiffness function and the values shown in Table 2. Running Simulation After the similarity between actuator simulation and actuator prototype was established. Figure 8 shows the response of both θ1 and θ2 . Although this test does not provide great insight into the performance of the actuator. and we speculate that the most likely reason for this discrepancy is an inaccurate model of friction and stiction. we tested the simulation in a running application.Motor Position (rad) 15 10 5 0 −5 −1 1 −0. Figure 9 shows the resulting Bode plots. Comparison of the position transfer function (ratio of motor position to leg position) for the physical and simulated actuators. A simulated runner was created using SD/FAST2 . C. 10.00474 m 0. the behavior does not change with frequency. at three different values of Keff .00134 kg·m2 0.6 kg 8. for both the simulation and the actuator. as input and leg position.5 0 −0. with different values for K ∗ . Keff = 5000 10 Magnitude(dB) 5 0 −5 −10 0 10 100 Phase (deg) 50 0 −50 −100 0 10 1 2 3 1 2 3 Fig. 9. Bode plots of the actuator gain were created with motor position. and one tuned to the same value.1 m Fig. creating basic 2-link legs with a knee and hip joint attached to a body. was commanded to track a sine wave beginning at a frequency of 1 Hz and cycling through to 6 Hz. Dynamic Actuator Simulation A dynamic simulation of the actuator prototype was created using the rotational model shown in Figure 3. x2 . 8.5 −1 −0. For the tuned system with natural dynamics matching the commanded stiffness. and ParametricTechnology Corporation is a registered trademark of SolidWorks Corporation . its response was tested with a step-input on the spring set point and compared to the same input on the physical actuator. Measured and simulated actuator position response to a step command in spring rest length K* = 5000 K* = 3000 K* = 7000 B. and the position of x2 was recorded. The actuator stiffness Keff is a constant for the three plots. As the frequency increases.085 kg·m2 0. An ideal spring was placed at each knee. Inc.38 N·m·s/rad variable M1 = M2 = r1 r2 J1 2 r1 J2 2 r2 10 10 10 10 10 Frequency (Hz) 10 value 59. and from several system identification tests. lower. For all plots. it is apparent from the figure that the phase begins to lag or lead. Running controllers were 2 SD/FAST is a trademark of Symbolic Dynamics. The values in Table 2 were calculated from the SolidWorks1 design of the actuator. The values do not match exactly. After adding these values to the computer simulation. one set higher than Keff . while the exhibited stiffness begins to diverge from the commanded stiffness K ∗ .5 0 Time (s) 0.5 1 1.

and the desired spring function. The simulated actuator has a slightly lower hopping height and resulting shorter stride length.5 1 1. 11. “Mechanics of locomotion. pp. pp. pp. H. Salisbury. the necessary rate of stiffness adjustment. Farley. 1986. Schwind and D. R. Winters and P. pp. Mechanisms. [4] R. and T. 65–78. [13] D. 181–186. Heglund. “The spring-mass model for running and hopping. 1997.” The International Journal of Robotics Research. due to the motor inertia M1 . 2003. For example. Koditschek. A. McMahon and P.5 3 Fig. Crago. P. D. “The sources of external work in level walking and running. T. 989–993. The actuator design presented here has mechanical energy storage. J. vol. 1970. A. Albequerque.” in Proceedings of the 1997 IEEE International Conference on Robotics and Automation. A.” Journal of Experimental Biology. 265. H. Louie.” Journal of Applied Physiology. 2000. tunable compliance. a discrete change to the commanded spring set point does not cause an instantaneous change the actual spring set point. as shown in Figure 11(b). while variable compliance is a useful control parameter for SLIP model running. “The role of compliance in mammalian running gaits. Famini. [14] T. R.” in Biomechanics and Neural Control of Posture and Movement.” Journal of Applied Physiology. 5. Eds. Taylor. 893–904.a comparative approach to legged locomotion. [7] G. McMahon and G. “Similarity in multilegged locomotion: Bouncing like a monopode. 1044–1055. pp. A. Cavagna. After tuning the controller and adding user-inputs to change running speed and hopping height. vol. 1390–1395. Soc. Ferris and C. no. Figure 12 shows the vertical position of the center of mass over time. [17] C. depicted as M1 in Figure 11(a). Equivalent SLIP models. Zamboni.5 1 0. McMahon. no. 422. q (b) Rotational model Fig. R243– R261. [10] T. vol. Legged Robots That Balance. Farley. 1984. “Leg stiffness and stride frequency in human running. J. 1985. C. V. There are still many unanswered questions for further investigation. Lair. “Mechanism of leg stiffness adjustment for hopping on surfaces of different stiffnesses. K. McMahon. T. 1. McMahon. C ONCLUSIONS Physical compliance and mechanical energy storage are crucial for a successful running gait. ch. [8] G. vol. 3. [15] T. “The influence of track compliance on running. vol. and easily accounted for in future controller revisions. This is because. McGraw Hill.” in Proc. pp. Valiant. J. Strien.” Journal of Physiology. C. 3.” Journal of Biomechanics.5 0 0 Ideal Knee Spring Simulated Actuator Height of CoM (m) 0. 1965. and E. “The mechanics of running: How does stiffness couple with speed?” Journal of Biomechanics. H. E. Raibert. Gonzalez.. R. [22] M. [9] G. pp. “Groucho running. A. T. V. Glasheen. J. pp. 1989. K. and zero backlash. no. A.” in IEEE Transactions on Robotics and Automation. 262. Within its range of physical compliance. this discrepancy is relatively minor. 639–657. P. and C. we would like to determine what stiffness range is required for robust running on various surfaces. pp. H. vol. Full. C. pp. “Mechanical work in terrestrial locomotion: Two basic mechanisms for minimizing energy expenditure. 11/12. “Characterization of monopod equilibrium gaits. 29.” Journal of Experimental Biology. and R. 509–517. M. A. 1989. Mass.” Journal of Biomechanics. 2326–2337. “Mechanical bandwidth as a guidline to high-performance manipulator design. Cheng. 263–282. Cavagna. pp. New Mexico. 29. Farley. vol. pp. ACKNOWLEDGEMENTS This work is supported in part by an NSF Graduate Fellowship held by the first author and by the Robotics Institute of Carnegie Mellon University R EFERENCES [1] W. Linkages. Leg stiffness was held constant. McMahon. [3] T. 493–494. However. The primary difference between this model and the original SLIP shown in Figure 2 is the presence of the reflected motor inertia. “Running springs: Speed and animal size. 2.” Journal of Biomechanics. P. Thys. 1995. [21] W.: MIT Press. 1979. vol. the runner maintained a stable gait within reasonable bounds. E. [11] D. [18] C. no. C. 12. We then replaced the ideal spring at each knee with the full dynamic simulation of the actuator. pp. 22. 1977. pp. Lond. [2] R. 1991. R. A. [20] N. and C. “Running in the real world: adjusting leg stiffness for different surfaces. 1987. P. 185. R. Louie. no. N. no. J. 1217–1227. 1990. and calculate the leg touchdown angle and energy insertion required to stabilize a running gait at a certain hopping height. 2. January 1998. Blickhan. 62. T. Cambridge. “Musculoskeletal dynamics in rhythmic systems . M.” Journal of Comparative Physiology. unlike an ideal spring. vol. Cavagna. 1993. Greene. pp. vol. Kram. “Are fast-moving elephants really running?” Nature. caused by lower energy insertion with each hop. pp. G. 1998. vol. Blickhan and R. 1986–1992. the actuator has virtually no bandwith limitations and is similar to an ideal SLIP. Comparison of hopping behavior between ideal knee spring and realistic actuator model in hopping simulation adapted from [22].” The American Physiological Society. Chironis. Townsend and J. 12. T. vol. Houdijk. incorporating actuator models as shown in Figure 3. 23. T. April 1997. [12] C.” in IEEE. [5] R. Types of Noncircular Gears. Full and C. both linear and rotational. 3. 1993. New York: Springer-Verlag. [19] J. T. 71–87. 4–28.5 Time (s) 2 2. pp. [6] J. 1976.M X M1’ M K q (a) Linear model prototype could result in an effective actuation method for highly dynamic legged locomotion. 241–244. the concepts embodied in this actuator . 7. Raibert. 15–22.” American Journal Physiology. “Adjusting step length for rough terrain. 233. Ferris. Farley. “Interaction of leg stiffness and surface stiffness during human hopping. “Elastic bounce of the body. pp. Farley. 279–282. low friction. Frederick. Farley and O. 115. Based on the results presented. pp. A. no. Hodgins and M. Hutchinson. and A. [16] T. and M. vol.” The American Physiological Society. and Mechanical Controls.

Sign up to vote on this title
UsefulNot useful