University of Pennsylvania ScholarlyCommons Center for Human Modeling and Simulation Department of Computer & Information Science March 1996 Animating Human Locomotion with Inverse Dynamics Norman I. Badler University of Pennsylvania, [email protected] Hyeongseok Ko Seoul National University Follow this and additional works at: https://repository.upenn.edu/hms Recommended Citation Badler, N. I., & Ko, H. (1996). Animating Human Locomotion with Inverse Dynamics. Retrieved from https://repository.upenn.edu/hms/38 Copyright 1996 IEEE. Reprinted from IEEE Computer Graphics and Application, Volume 6, Issue 2, March 1996, pages 50-59. Publisher URL: http://dx.doi.org/10.1109/38.486680 This material is posted her with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to [email protected]. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. This paper is posted at ScholarlyCommons. https://repository.upenn.edu/hms/38 For more information, please contact [email protected]. Animating Human Locomotion with Inverse Dynamics Abstract Locomotion is a major component of human activity, and there have been many attempts to reveal its principles through the application of physics and dynamics. Both computer graphics and robotics continue such efforts, but many problems remain unsolved, even in characterizing the simplest case: linear, forward, rhythmic walking. Comments Copyright 1996 IEEE. Reprinted from IEEE Computer Graphics and Application, Volume 6, Issue 2, March 1996, pages 50-59. Publisher URL: http://dx.doi.org/10.1109/38.486680 This material is posted her with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to [email protected]. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. This journal article is available at ScholarlyCommons: https://repository.upenn.edu/hms/38 Animating Gait and Locomotion Animating Human Locomotion with Hyeongseok KO Seou t Fiationat University ocomotion is a major component of human motion and to maintain the joint torques within a mod: Lactivity, and there have been many eratei range imposed by human strength limits. This attempts to reveal its principles through the application method corrects or predicts a motion as indicated by the of physics and dynamics. Both computer graphics and inverse dynamics analysis. robotics continue such efforts, but many problems Dynamic correctness is a sufficientcondition for real- remain unsolved, even in characterizing the simplest istic imotion of nonliving objects. In animating a self- case: linear, forward, rhythmic walking. actuated system, however, visual realism is another Since all mechanical linkage important, separate criterion for determining the suc- systems are subject to forces and cess of ,a technique. Dynamic correctness is not a suffi- This technique generates physical laws, the computation of cient condition for this visual realism. An animation of dynamics seems a promising ap- dynamically balanced walking that is also “comfortable” dynamically sound walking proach to humanmotion problems. in the sense of avoiding strengthviolations can still look Forward dynamics has proved use- quite different from normal human walking. In this arti- motion for any human gait, ful in predicting the motion of non- cle, a visually realistic and dynamically sound anima- living objects. For example, you can tion of human locomotion is obtained using an effective figure scale, or motion path simulate a swinging chain bydefin- combination of kinematic and dwamic techniques. ing the initial state and then just by maintaining balance and integrating the effect of gravityor Relalted work other forces acting on the system. Boulic et a1.l and KO and Badle~?~attempted kine- keeping joint stress within Unfortunately, the theory is less matic gleneralizationof empirical walking data to gen- successful in animating the move- erate lolcomotion along a curved path and intermittent, the torque given by ments of “self-actuated’’ systems, nonrhythmic stepping in any direction (forward, back- namely, living creatures. Because ward, lateral, clockwise, and counterclockwise). Their empirical strength data. the major force components-the kinematic generations, however, do not handle the internal muscular forces and important case of a load or force attached to the body. torques-are not known a priori over time, you cannot Brudierlin and Calvert’ used a combination of kine- use forward dynamics to predict how the human body matic considerations and dynamic motion control for will walk. Nor is there any known physical law to pre- goal-directed animation of human walking. For biped dict how that walk will change if an external force acts running, Girard6computed the impulses at each liftoff on the model. Accordingly, it is not easy to guess the joint that clrive the center of mass along the given path. He torque patterns that will drive the model to take a step. also added banking, which is a function of the velocity Even if it takes a step, the result is unlikely to resemble and curvature of running, for dynamic stability. a human walking pattern. Inroblotics, many researchers have built actual bipedal The alternative to the apparentlyintractable problem walking robot^,^ but walking stability has not been easy of specifying the joint torque patterns in advance is to to achieve. The gait pattern for walking robots is pre- use inverse dynamics to analyze the torques and forces ” designed off line, and robot makers are not usuallycon- required for the given motion. Such an analysis can cerneld with achieving human-like realism in stepping. show, for example, that the motion induces excessive Rather, they focus on the stability of the whole system. torque, that the system is out of balance at a certain point, or that the step length is too great. In this article, Overviiew of the Speedy system we present a method of using an inverse dynamics com- In conventional analysis systems like the Dynamic putation to dynamically balance the resulting walking Analysis and Design System from Computer Aided Design 50 March 1996 0272-1 7-1 6/96/85 00 0 1996 IEEE Software (Iowa City, Iowa), the analysis and correction p, phases are temporally disjoint: A given motion is first ana- Kinematic lyzed (for example, to obtain the joint torques) and the rl locomotion I generation: eo\ required correctionis then computed. This works fine for 1 mechanical linkagesin non-self-actuatedsystems; but in 1 Overviewof self-actuatedsystems, the correctioncan make the analy- the locomotion sis result incorrect and thus necessitate analysis of the control. new motion to ensure its dynamic soundness. Thus, gen- Inverse dynamics erating a stable gait can involve many iterations. Balance control To avoid the problem, we built a real-time inverse dynamics package and real-time motion corrector. These are used in a single-phase dynamic motion gen- erator called Speedy. It calls the analyzer and corrector alternately for each frame to generate dynamically The primary parameters driving the input to the Klog sound motion in real time. consist of the next footprint and the designation of the The Speedy system controls human locomotion so stepping foot. This footprint-driven approach provides that balance is maintained and joint stress is kept with- high-fidelity control of human locomotion. The prima- in the available torque given by empirical strength data. ry parameters can be supplied from many so~rces:~reac- The strength data can be used to simulate comfortable tive or nonreactive path planning systems, virtual reality walking or fatigued walking according to a scaled set of applications, or interactivelyconstructed spline curves, available torques. and so forth. (Note that there are a few other ways of Figure 1gives an overview of the Speedy system. A specifyinglocomotion, such as the path of the center of kinematic locomotion generator (“Klog”) spawns walk- mass or pelvis. However, these paths do not have exact ing motions, which are then analyzed and dynamically control of the footprints, so the agent can fail, for exam- modified to meet balance and stress constraints by the ple, to step over a pit on the way. In comfort control, the Dyncontrol module. locomotion path is sometimesgiven as the primary para- We use @(t)to denote all the joint angles and the posi- meter, which allows taking smaller steps to resolve a tion of the figure base in world coordinates at frame t. To strengthviolation. Also, in balance control the locomo- control inverse dynamics requires an underlying motion tion path is sometimes the primary parameter to allow 00and some mechanism for modifying it over time. We narrowing of the lateral step width and thus to reduce use the normal gait pattern with no load for 00.The body swaying.) modificationis done through the control parameters A0 Klog is goal driven in that the resulting motion predicted through inverse dynamics. Conceptually,we achieves the goal footprint very accurately. Even though can write the modification as some other aspects of the walking motion may be altered by later application of Dyncontrol, the goal achievement is not affected. Klog has two phases. During the first phase it pre- At each frame, real-time inverse dynamics provide computes the kinematic motion assuming no load or balance information and exerted force and torque at external force is attached.