Bio-Inspired Controller for a Robot Cheetah with a Neural Mechanism Controlling Leg Muscles

The realization of a high-speed running robot is one of the most challenging problems in developing legged robots.The excellent performance of cheetahs provides inspiration for the control and mechanical design of such robots.This paper presents a three-dimensional model of a cheetah that predicts the locomotory behaviors of a running cheetah.Applying biological knowledge of the neural mechanism,we control the muscle flexion and extension during the stance phase,and control the positions of the joints in the flight phase via a PD controller to minimize complexity.The proposed control strategy is shown to achieve similar locomotion of a real cheetah.The simulation realizes good biological properties,such as the leg retraction,ground reaction force,and spring-like leg behavior.The stable bounding results show the promise of the controller in high-speed locomotion.The model can reach 2.7 m·s- 1 as the highest speed,and can accelerate from 0 to 1.5 m·s -1 in one stride cycle.A mechanical structure based on this simulation is designed to demonstrate the control approach,and the most recently developed hindlimb controlled by the proposed controller is presented in swinging-leg experiments and jump-force experiments.     

A simple method for measuring stiffness during running

The spring-mass model, representing a runner as a point mass supported by a single linear leg spring, has been a widely used concept in studies on running and bouncing mechanics. However, the measurement of leg and vertical stiffness has previously required force platforms and high-speed kinematic measurement systems that are costly and difficult to handle in field conditions. We propose a new "sine-wave" method for measuring stiffness during running. Based on the modeling of the force-time curve by a sine function,this method allows leg and vertical stiffness to be estimated from just a few simple mechanical parameters: body mass, forward velocity, leg length, flight time, and contact time. We compared this method to force-platform-derived stiffness measurements for treadmill dynamometer and overground running conditions, at velocities ranging from 3.33 m.s-1 to maximal running velocity in both recreational and highly trained runners. Stiffness values calculated with the proposed method ranged from 0.67 % to 6.93 % less than the force platform method, and thus were judged to be acceptable. Furthermore, significant linear regressions (p < 0.01) close to the identity line were obtained between force platform and sine-wave model values of stiffness. Given the limits inherent in the use of the spring-mass model, it was concluded that this sine-wave method allows leg and stiffness estimates in running on the basis of a few mechanical parameters, and could be useful in further field measurements.     

Running in the real world: adjusting leg stiffness for different surfaces.

A running animal coordinates the actions of many muscles, tendons, and ligaments in its leg so that the overall leg behaves like a single mechanical spring during ground contact. Experimental observations have revealed that an animal''s leg stiffness is independent of both speed and gravity level, suggesting that it is dictated by inherent musculoskeletal properties. However, if leg stiffness was invariant, the biomechanics of running (e.g. peak ground reaction force and ground contact time) would change when an animal encountered different surfaces in the natural world. We found that human runners adjust their leg stiffness to accommodate changes in surface stiffness, allowing them to maintain similar running mechanics on different surfaces. These results provide important insight into mechanics and control of animal locomotion and suggest that incorporating an adjustable leg stiffness in the design of hopping and running robots is important if they are to match the agility and speed of animals on varied terrain.     

Leg stiffness in human running: Comparison of estimates derived from previously published models to direct kinematic-kinetic measures

It is not presently clear whether mathematical models used to estimate leg stiffness during human running are valid. Therefore, leg stiffness during the braking phase of ground contact of running was calculated directly using synchronous kinematic (high-speed motion analysis) and kinetic (force platform) analysis, and compared to stiffness calculated using four previously published kinetic models. Nineteen well-trained male middle distance runners (age=21.1±4.1yr; VO(2max)=69.5±7.5mlO(2)kg(-1)min(-1)) completed a series of runs of increasing speed from 2.5 to 6.5ms(-1). Leg stiffness was calculated directly from kinetic-kinematic analysis using both vertical and horizontal forces to obtain the resultant force in the line of leg compression (Model 1). Values were also estimated using four previously published mathematical models where only force platform derived and anthropometric measures were required (Models 2-5; Morin et al., 2005, Morin et al., 2011, Blum et al., 2009, Farley et al., 1993, respectively). The greatest statistical similarity between leg stiffness values occurred with Models 1 and 2. The poorest similarity occurred when values from Model 4 were compared with Model 1. Analyses suggest that the poor correlation between Model 1 other models may have resulted from errors in the estimation in change in leg length during the braking phase. Previously published mathematical models did not provide accurate leg stiffness estimates, although Model 2, used by Morin et al. (2005), provided reasonable estimates that could be further improved by the removal of systematic error using a correction factor (K=1.0496K(Model2)).     

Task-Level Strategies for Human Sagittal-Plane Running Maneuvers Are Consistent with Robotic Control Policies

The strategies that humans use to control unsteady locomotion are not well understood. A “spring-mass” template comprised of a point mass bouncing on a sprung leg can approximate both center of mass movements and ground reaction forces during running in humans and other animals. Legged robots that operate as bouncing, “spring-mass” systems can maintain stable motion using relatively simple, distributed feedback rules. We tested whether the changes to sagittal-plane movements during five running tasks involving active changes to running height, speed, and orientation were consistent with the rules used by bouncing robots to maintain stability. Changes to running height were associated with changes to leg force but not stance duration. To change speed, humans primarily used a “pogo stick” strategy, where speed changes were associated with adjustments to fore-aft foot placement, and not a “unicycle” strategy involving systematic changes to stance leg hip moment. However, hip moments were related to changes to body orientation and angular speed. Hip moments could be described with first order proportional-derivative relationship to trunk pitch. Overall, the task-level strategies used for body control in humans were consistent with the strategies employed by bouncing robots. Identification of these behavioral strategies could lead to a better understanding of the sensorimotor mechanisms that allow for effective unsteady locomotion.     

Mechanical power in running: a comparison of different approaches

The purposes of the present study were: (1) to compare four different methods of calculating mechanical power in running on the basis of comparable data over a wide range of running velocity; (2) to examine the linearity of the relation between mechanical power as calculated with the four methods and running velocity. Eight runners participated in the investigation (height: 1.82 +/- 0.03 m, body mass: 81.05 +/- 4.69 kg). A Kistler force platform registered all components of the ground reaction force (1000 Hz) during one foot ground contact, which was additionally video taped using two high-speed video cameras running at 120 Hz. Four different methods were used to calculate mechanical power. Two methods determined the mechanical power due to the work done on the athletes' center of mass and two were calculated from the motion of the athletes' segments. The four different methods provided different relations between mechanical power and running velocity. The calculations on the basis of kinematic data cannot be recommended to determine efficiency of movement. The methods based on ground reaction force measurements revealed significant linear relations (r = 0.90, r2 = 0.84) between running velocity and mechanical power.     

Consequences of forward translation of the point of force application for the mechanics of running

In running humans, the point of force application between the foot and the ground moves forwards during the stance phase. Our aim was to determine the mechanical consequences of this 'point of force translation' (POFT). We modified the planar spring-mass model of locomotion to incorporate POFT, and then compared spring-mass simulations with and without POFT. We found that, if leg stiffness is adjusted appropriately, it is possible to maintain very similar values of peak vertical ground reaction force (GRF), stance time, contact length and vertical centre of mass displacement, whether or not POFT occurs. The leg stiffness required to achieve this increased as the distance of POFT increased. Peak horizontal GRF and mechanical work per step were lower when POFT occurred. The results indicate that the lack of POFT in the traditional spring-mass model should not prevent it from providing good predictions of peak vertical GRF, stance time, contact length and vertical centre of mass displacement in running humans, if an appropriate spring stiffness is used. However, the model can be expected to overestimate peak horizontal GRF and mechanical work per step. When POFT occurs, the spring stiffness in the traditional spring-mass model is not equivalent to leg stiffness. Therefore, caution should be exercised when using spring stiffness to understand how the musculoskeletal system adapts to different running conditions. This can explain the contradictory results in the literature regarding the effect of running speed on leg stiffness.     

Distributed mechanical feedback in arthropods and robots simplifies control of rapid running on challenging terrain

Terrestrial arthropods negotiate demanding terrain more effectively than any search-and-rescue robot. Slow, precise stepping using distributed neural feedback is one strategy for dealing with challenging terrain. Alternatively, arthropods could simplify control on demanding surfaces by rapid running that uses kinetic energy to bridge gaps between footholds. We demonstrate that this is achieved using distributed mechanical feedback, resulting from passive contacts along legs positioned by pre-programmed trajectories favorable to their attachment mechanisms. We used wire-mesh experimental surfaces to determine how a decrease in foothold probability affects speed and stability. Spiders and insects attained high running speeds on simulated terrain with 90% of the surface contact area removed. Cockroaches maintained high speeds even with their tarsi ablated, by generating horizontally oriented leg trajectories. Spiders with more vertically directed leg placement used leg spines, which resulted in more effective distributed contact by interlocking with asperities during leg extension, but collapsing during flexion, preventing entanglement. Ghost crabs, which naturally lack leg spines, showed increased mobility on wire mesh after the addition of artificial, collapsible spines. A bioinspired robot, RHex, was redesigned to maximize effective distributed leg contact, by changing leg orientation and adding directional spines. These changes improved RHex's agility on challenging surfaces without adding sensors or changing the control system.     

Metabolic energy and muscular activity required for leg swing in running.

The metabolic cost of leg swing in running is highly controversial. We investigated the cost of initiating and propagating leg swing at a moderate running speed and some of the muscular actions involved. We constructed an external swing assist (ESA) device that applied small anterior pulling forces to each foot during the first part of the swing phase. Subjects ran on a treadmill at 3.0 m/s normally and with ESA forces up to 4% body weight. With the greatest ESA force, net metabolic rate was 20.5% less than during normal running. Thus we infer that the metabolic cost of initiating and propagating leg swing comprises approximately 20% of the net cost of normal running. Even with the greatest ESA, mean electromyograph (mEMG) of the medial gastrocnemius and soleus muscles during later portions of stance phase did not change significantly compared with normal running, indicating that these muscles are not responsible for the initiation of leg swing. However, with ESA, rectus femoris mEMG during the early portions of swing phase was as much as 74% less than during normal running, confirming that it is responsible for the propagation of leg swing.     

Mechanical analysis of the landing phase in heel-toe running.

Results of mechanical analyses of running may be helpful in the search for the etiology of running injuries. In this study a mechanical analysis was made of the landing phase of three trained heel-toe runners, running at their preferred speed and style. The body was modeled as a system of seven linked rigid segments, and the positions of markers defining these segments were monitored using 200 Hz video analysis. Information about the ground reaction force vector was collected using a force plate. Segment kinematics were combined with ground reaction force data for calculation of the net intersegmental forces and moments. The vertical component of the ground reaction force vector Fz was found to reach a first peak approximately 25 ms after touch-down. This peak occurs because, in the support leg, the vertical acceleration of the knee joint is not reduced relative to that of the ankle joint by rotation of the lower leg, so that the support leg segments collide with the floor. Rotation of the support upper leg, however, reduces the vertical acceleration of the hip joint relative to that of the knee joint, and thereby plays an important role in limiting the vertical forces during the first 40 ms. Between 40 and 100 ms after touch-down, the vertical forces are mainly limited by rotation of the support lower leg. At the instant that Fz reaches its first peak, net moments about ankle, knee and hip joints of the support leg are virtually zero. The net moment about the knee joint changed from -100 Nm (flexion) at touch-down to +200 Nm (extension) 50 ms after touch-down. These changes are too rapid to be explained by variations in the muscle activation levels and were ascribed to spring-like behavior of pre-activated knee flexor and knee extensor muscles. These results imply that the runners investigated had no opportunity to control the rotations of body segments during the first part of the contact phase, other than by selecting a certain geometry of the body and muscular (co-)activation levels prior to touch-down.     

Optimum walking techniques for quadrupeds and bipeds

A new theory is presented which describes quadrupedal as well as bipedal walking. It avoids errors which occurred in previous theories by evaluating separately the work done by each leg instead of deriving net work from mechanical energy fluctuations. It takes particular account of two parameters, the duty factor β (the fraction of the stride for which each foot is on the ground) and a parameter q which defines the time course of the force on each foot. It shows that for any given speed there is an optimum (β, q ) which minimizes the energy cost of locomotion. These (β, q ) are only a little different for bipeds and quadrupeds except near the critical speed at which the optimum moves abruptly from walking (high β) to running (low β). Walking men use (β, q ) close to the theoretical optima, but with slightly higher q. Walking dogs and sheep use q which are lower than the optimum values except at very low speeds. Some of the energy which would otherwise be required for locomotion may be saved by storage of elastic strain energy in tendons etc. This mechanism is more effective in running than in fast walking, which may be why men change from walking to running at lower speeds than the inelastic theory suggests.     

Structural Achilles tendon properties in athletes subjected to different exercise modes and in Achilles tendon rupture patients.

The prevalence of Achilles tendon (AT) injury is high in various sports, and AT rupture patients have been reported to have a 200-fold risk of sustaining a contralateral rupture. Tendon adaptation to different exercise modes is not fully understood. The present study investigated the structural properties of the AT in male elite athletes that subject their AT to different exercise modes as well as in Achilles rupture patients. Magnetic resonance imaging of the foot and leg, anthropometric measurements, and maximal isometric plantar flexion force were obtained in 6 male AT rupture patients and 25 male elite athletes (kayak/control group n = 9, volleyball n = 8 and endurance running n = 8). AT cross-sectional area (CSA) was normalized to body mass. Runners had a larger normalized AT CSA along the entire length of the tendon compared with the control group (P < 0.05). The volleyball subjects had a larger normalized CSA compared with the control group (P < 0.05) in the area of thinnest tendon CSA. No structural differences of the AT were found in the rupture subjects compared with the control group. Rupture subjects did not subject their AT to greater force or stress during a maximal voluntary isometric plantar flexion than the other groups. The CSA of the triceps surae musculature was the strongest predictor of AT CSA (r(s) = 0.569, P < 0.001). This study is the first to show larger CSA in tendons that are subjected to intermittent high loads. AT rupture patients did not display differences in structural or loading properties of the tendons.     

Control of constraint forces and trajectories in a rich sensory and actuation environment

A simple control strategy is proposed and applied to a class of non-linear systems that have abundant sensory and actuation channels as in living systems. The main objective is the independent control of constrained trajectories of motion, and control of the corresponding constraint forces. The peripheral controller is a proportional, derivative and integral (PID) controller. A central controller produces, via pattern generators, reference signals that are the desired constrained position and velocity trajectories, and the desired constraint forces. The basic tenet of the this hybrid control strategy is the use of two mechanisms: 1. linear state and force feedback, and 2. non-linear constraint velocity feedback - sliding mode feedback. The first mechanism can be envisioned as a high gain feedback systems. The high gain attribute imitates the agonist-antagonist co-activation in natural systems. The strategy is applied to the control of the force and trajectory of a two-segment thigh-leg planar biped leg with a mass-less foot cranking a pedal that is analogous to a bicycle pedal. Five computational experiments are presented to show the effectiveness of the strategy and the performance of the controller. The findings of this paper are applicable to the design of orthoses and prostheses to supplement functional electrical stimulation for support purposes in the spinally injured cases.     

Biological Jumping Mechanism Analysis and Modeling for Frog Robot

This paper presents a mechanical model of jumping robot based on the biological mechanism analysis of frog. By biological observation and kinematic analysis the frog jump is divided into take-offphase, aerial phase and landing phase. We find the similar trajectories of hindlimb joints during jump, the important effect of foot during take-off and the role of forelimb in supporting the body. Based on the observation, the frog jump is simplified and a mechanical model is put forward. The robot leg is represented by a 4-bar spring/linkage mechanism model, which has three Degrees of Freedom （DOF） at hip joint and one DOF （passive） at tarsometatarsal joint on the foot. The shoulder and elbow joints each has one DOF for the balancing function of arm. The ground reaction force of the model is analyzed and compared with that of frog during take-off. The results show that the model has the same advantages of low likelihood of premature lift-off and high efficiency as the frog. Analysis results and the model can be employed to develop and control a robot capable of mimicking the jumping behavior of frog.     

Stance phase control of above-knee prostheses: knee control versus SACH foot design

The mobility of above-knee amputees (A/K) is limited, in part, due to the performance of A/K prostheses during the stance phase. Currently stance phase control of most conventional A/K prostheses can only be achieved through leg alignment and choice of the SACH (Solid Ankle Cushioned Heel) foot. This paper examines the role of the knee controller in relation to a SACH foot during the stance phase of level walking. The three-dimensional gait mechanics were measured under two stance phase conditions. In the first set of trials, the amputee used a prosthesis with a conventional knee controller that allowed the amputee to maintain the knee joint in full extension during the stance phase. In the second set of trials, the prosthetic knee, during stance, echoed the modified kinematics of the amputee's sound (intact) knee that had been recorded during the previous sound stance phase. Analysis and interpretation of the data indicate the following: (1) SACH foot design can strongly influence the walking mechanics independent of the knee controller; (2) knee controller design and SACH foot design are mutually interdependent; and (3) normal kinematics imposed on the prosthetic knee does not necessarily produce normal hip kinematics (e.g. reduce the abnormal rise in the prosthetic side hip trajectory). Future research is necessary to explore and exploit the interdependency of prosthetic knee control and foot design.     

Running on an incline.

Seven male subjects ran at 3.0 m/s on a motorized treadmill including a force platform under the tread. The subjects ran at each of five treadmill inclinations: +0.17, +0.077, 0, -0.077, and -0.17 radians. The position of the subjects' legs were read from ciné films (100 frames/s). Results of the film and force plate analysis generally corroborated the "hanging triangle" hypothesis, which postulates that the angle between the leg and the vertical upon foot strike does not change as the treadmill is tipped up or down. A mathematical model of running, in which the leg is represented as a nonlinear spring, made satisfactory predictions of the way many parameters of running change with the treadmill angle, including the length of the leg at touchdown and liftoff and the peak leg force in the middle of a step. The peak leg force reaches a maximum at a treadmill angle near -0.12 radians, close to the downhill angle where other authors have found a minimum in the rate of oxygen consumption.     

Three-dimensional measurement of rearfoot motion during running

Excessive ranges of motion during running have been speculated to be connected to injuries to the lower extremities. Movement of the foot and lower leg has commonly been studied with two-dimensional techniques. However, differences in the alignment of the longitudinal axis of the foot with the camera axis will produce measurement errors for projected angles of the lower extremities. A three-dimensional approach would not have this limitation. The purpose of this study is to present a three-dimensional model for calculation of angles between lower leg and foot, lower leg and ground, and foot and ground, and to compare results from treadmill running derived from this model with results derived from a two-dimensional model for different alignment angles between foot axis and camera axis. A two camera Selspot system was used to obtain three-dimensional information on motion of the studied segments. It was found that several two-dimensional variables measured from a posterior view are very sensitive to the alignment angle between the foot and the camera axis. Some variables change as much as 1 degrees for every 2 degrees of change of the alignment angle. The large influence of rotations other than the measured one in two-dimensional measurements makes advisable the use of a three-dimensional model when studying motion between foot and lower leg during running.     

Mechanical energy and effective foot mass during impact loading of walking and running

The human heel pad is considered an important structure for attenuation of the transient force caused by heel-strike. Although the mechanical properties of heel pads are relatively well understood, the mechanical energy (Etot) absorbed by the heel pad during the impact phase has never been documented directly because data on the effective foot mass (Meff) was previously unavailable during normal forward locomotion. In this study, we use the impulse-momentum method (IMM) for calculating Meff from moving subjects. Mass-spring-damper models were developed to evaluate errors and to examine the effects of pad property, upper body mass, and effective leg spring on Meff. We simultaneously collected ground reaction forces, pad deformation, and lower limb kinematics during impact phase of barefoot walking, running, and crouched walking. The latter was included to examine the effect of knee angle on Meff. The magnitude of Meff as a percentage of body mass (M(B)) varies with knee angle at impact and significantly differs among gaits: 6.3%M(B) in walking, 5.3%M(B) in running, and 3.7%M(B) in crouched walking. Our modeling results suggested that Meff is insensitive to heel pad resilience and effective leg stiffness. At the instant prior to heel strike, Etot ranges from 0.24 to 3.99 J. The combination of video and forceplate data used in this study allows analyses of Etot and Etot as a function of heel-strike kinematics during normal locomotion. Relationship between Meff and knee angle provides insights into how changes in posture moderate impact transients at different gaits.     

Grip and limb force limits to turning performance in competition horses

Manoeuverability is a key requirement for successful terrestrial locomotion, especially on variable terrain, and is a deciding factor in predator-prey interaction. Compared with straight-line running, bend running requires additional leg force to generate centripetal acceleration. In humans, this results in a reduction in maximum speed during bend running and a published model assuming maximum limb force as a constraint accurately predicts how much a sprinter must slow down on a bend given his maximum straight-line speed. In contrast, greyhounds do not slow down or change stride parameters during bend running, which suggests that their limbs can apply the additional force for this manoeuvre. We collected horizontal speed and angular velocity of heading of horses while they turned in different scenarios during competitive polo and horse racing. The data were used to evaluate the limits of turning performance. During high-speed turns of large radius horizontal speed was lower on the bend, as would be predicted from a model assuming a limb force limit to running speed. During small radius turns the angular velocity of heading decreased with increasing speed in a manner consistent with the coefficient of friction of the hoof-surface interaction setting the limit to centripetal force to avoid slipping.     

Modeling muscle activity to study the effects of footwear on the impact forces and vibrations of the human body during running

A previously developed mass-spring-damper model of the human body is improved in this paper, taking muscle activity into account. In the improved model, a nonlinear controller mimics the functionality of the Central Nervous System (CNS) in tuning the mechanical properties of the soft-tissue package. Two physiological hypotheses are used to determine the control strategies that are used by the controller. The first hypothesis (constant-force hypothesis) postulates that the CNS uses muscle tuning to keep the ground reaction force (GRF) constant regardless of shoe hardness, wherever possible. It is shown that the constant-force hypothesis can explain the existing contradiction about the effects of shoe hardness on the GRF during running. This contradiction is emerged from the different trends observed in the experiments on actual runners, and experiments in which the leg was fixed and exposed to impact. While the GRF is found to be dependent on shoe hardness in the former set of experiments, no such dependency was observed in the latter. According to the second hypothesis, the CNS keeps the level of the vibrations of the human body constant using muscle tuning. The results of the study show that this second control strategy improves the model such that it can correctly simulate the effects of shoe hardness on the vibrations of the human body during running.