Motion Perturbation Based on Simple Neuromotor Control Models

KangKang Yin, Michael B. Cline, Dinesh K. Pai

Proceedings of the 11th Pacific Conference on Computer Graphics and Applications

Pages 445-449, 2003

Motivation

Motion capture is widely used for character animation. One major challenge of this technique is how to modify the captured motion in plausible ways. Existing techniques have focused on the physics of motion but can produce unnatural variations of motions.

Goal of This Research

The goal is to use a simple human neuromuscular control model to create animations that appear more natural than those that rely upon physics alone.

Goal of This Paper

This paper presents a way to incorporate a simple human neuromotor control model into dynamic simulation systems. This leads to a technique for modifying motion capture animations according to small unexpected disturbances in a dynamic environment in a way that makes the resultant motion still seem natural.

The motor control model used in this paper is a modification of an existing model containing an internal model of an object's inverse dynamics in the brain, which generates a feedforward motor command and a desired trajectory that is fed into a muscle-tendon system, which generates a force in response. This force, combined with effects from the environment, are sent to an inertial dynamics module that computes the actual trajectory, which is then sent back into the muscle-tendon system to form a feedback loop.

A matrix equation combining the Newton-Euler equations of motion with the constraint equations is used for both forward dynamics (where velocity is computed in terms of known muscle forces) and inverse dynamics (where the muscle forces are computed in terms of known accelerations). The motor control is integrated with dynamic simulation to create a two-stage algorithm:

1. Preprocessing: Use inverse dynamics to estimate feedforward torques from the motion capture data.
2. Forward Simulation: Compute muscle torques. These torques are a combination of the feedforward torques above and feedback torques computed based on the trajectories of objects in motion capture versus in dynamic simulation.

Related Work

Motor controllers for human dynamic simulation have been built by hand [12] and using machine learning techniques [7]. Adding motion capture into simulations makes the control problem easier to solve [25]. Spacetime constraints [24] [8] [19] put the motion editing problem into a constrained optimization framework by combining kinematic keyframing (space constraints) with dynamic simulation (time constraints). Motor learning techniques [20] [10] have led to the development of motor controllers for basic tasks such as locomotion.

The internal model theory states that the brain needs to learn an inverse dynamics model of an object to be controlled through motor learning. Then motor control is executed using feedforward muscle forces [15] [17]. Speed and accuracy are lower during motor learning than during well-trained movements due to the lack of good internal models [9].

Intrinsic mechanical properties of muscle and tendon produce proportional (stiffness) and derivative (viscosity) feedback forces without delay [11]. Muscle stiffness increases nonlinearly with generated force [23]. A simple model of this relationship is the bilinear model of muscle impedance [13] [22]. The motor control model used in this paper is from [14] [21]. Work similar to that in this paper appared in [26], but they used a high-gain feedback controller whereas this paper uses a low-gain feedback controller with feedforward control for most of the work, as is more realistic for biological systems.

General-purpose rigid body simulations [4] [5] are extended for motor control in this paper. Lagrange multipliers are used to compute constraint forces [3]. Details are given in [6]. Drift at joints caused by numerical error in forward dynamics is countered by a post-step stabilization scheme [2] [5]. A linearly implicit time stepping method is used for numerical integration [1]. Gradients of stiff forces are computed using automatic differentiation techniques [18]. Stiffness and damping constants for each joint angle have been measured in biomechanics [16].

Results

A skeleton model with 54 degrees of freedom was perturbed in two different experiments: (1) captured arm motions and (2) full body motions from football games. The skeleton responded to external disturbances and restored itself to the original motion naturally.

What This Paper Contributes

This paper is the first to explicitly incorporate human neuromotor control models into a human simulation system.

What This Paper Does Not Contribute

The motor control model used in this paper is very simple: the slower feedback loops (spinal and supraspinal feedback) are missing. This limits the system to disturbances that do not endanger balance and are recoverable by muscle-tendon feedback.

Bibliography

1

M. Anitescu and F. Potra.
A time-stepping method for stiff multibody dynamics with contact and friction.
International Journal for Numerical Methods in Engineering, 55:753-784, 2002.

2

U. Ascher, H. Chin, L. Petzold, and S. Reich.
Stabilization of constrained mechanical systems with daes and invariant manifolds.
Journal of Mechanics of Structures and Machines, 23:135-158, 1995.
Summary available .

3

D. Baraff.
Linear-time dynamics using lagrange multipliers.
Computer Graphics (SIGGRAPH), pages 137-146, 1996.
Summary available .

4

M. Cline.
Rigid body simulation with contact and constraints.
Master's thesis, University of British Columbia, 2002.

5

M. Cline and D. Pai.
Post-stabilization for rigid body simulation with contact and constraints.
In Proceedings of the 2003 IEEE International Conference on Robotics and Automation (ICRA 2003), pages 3744-3751, 2003.

6

M. Cline, K. Yin, and D. Pai.
Motion perturbation based on simple neuromotor control models.
Technical Report TR-2002-03.

7

P. Faloutsos, M. van de Panne, and D. Terzopoulos.
Composable controllers for physics-based character animation.
In Proceedings of SIGGRAPH 2001, pages 251-260, 2001.

8

M. Gleicher.
Retargetting motion to new characters.
In Proceedings of SIGGRAPH 1998, pages 33-42, 1998.

9

H. Gomi and M. Kawato.
Human arm stiffness and equilibrium-point trajectory during multi-joint movement.
Biological Cybernetics, 76:163-171, 1997.

10

R. Grzeszczuk, D. Terzopoulos, and G. Hinton.
Neuroanimator: Fast neural network emulation and control of physics-based models.
In Proceedings of SIGGRAPH 1998, pages 9-20, 1998.

11

S. Hall.
Basic Biomechanics.
Mosby, second edition, 1998.

12

J. Hodgins, W. Wooten, D. Brogan, and J. O'Brien.
Animating human athletics.
In Proceedings of SIGGRAPH 1995, pages 71-78, 1995.

13

N. Hogan.
Mechanical impedance of single and multi-articular systems.
In J. Winters and S. Woo, editors, Multiple Muscle Systems: Biomechanics and Movement Organization, chapter 9, pages 149-164. Springer-Verlag, 1990.

14

E. R. Kandel, J. H. Schwartz, and T. M. Jessell.
Principles of Neural Science.
McGraw-Hill, fourth edition, 2000.

15

M. Kawato.
Internal models for motor control and trajectory planning.
Current Opinion in Neurobiology, 9:718-727, 1999.

16

M. Latash and V. Zatsiorsky.
Joint stiffness: Myth or reality.
Human Movement Science, 12:653-692, 1993.

17

F. Mussa-Ivaldi.
Modular features of motor control and learning.
Current Opinion in Neurobiology, 9:713-717, 1999.

18

J. Nocedal and S. Wright.
Numerical Optimization.
Springer, 1999.

19

Z. Popovic and A. Witkin.
Physically based motion transformation.
Computer Graphics (SIGGRAPH), pages 11-20, 1999.
Summary available .

20

M. van de Panne and E. Fiume.
Sensor-actuator networks.
In Proceedings of SIGGRAPH 1993, pages 335-342, 1993.

21

T. Wang, G. Dordevic, and R. Shadmehr.
Learning the dynamics of reaching movements results in the modification of arm impedance and long-latency perturbation responses.
Biological Cybernetics, 85(6):437-448, 2001.

22

J. Winters and P. Crago, editors.
Biomechanics and Neural Control of Posture and Movement.
Springer-Verlag, 2000.

23

S. Wise and R. Shadmehr.
Motor Control.
Elsevier Science, 2002.

24

A. Witkin and M. Kass.
Spacetime constraints.
Computer Graphics (SIGGRAPH), 22:159-168, 1988.

25

V. Zordan and J. Hodgins.
Tracking and modifying upper-body human motion data with dynamic simulation.
In Computer Graphics Forum (Proceedings of Eurographics 1999), pages 13-22, 1999.

26

V. Zordan and J. Hodgins.
Motion capture-driven simulations that hit and react.
In ACM SIGGRAPH Symposium on Computer Animation, pages 89-96, 2002.