Reading Notes

Anderson F.C., Arnold A.S., Pandy M.G., Goldberg S.R., and Delp S.L. (in press). Simulation of Walking. In Rose, J. and Gamble, J.G. (eds.): Human Walking, 3rd Edition. Lippincott Williams & Williams, Inc., Philadelphia. 2005

http://nmbl.stanford.edu/publications/pdf/Anderson2006.pdf

Summary

Aims towards a dynamic simulation of walking that integrates facts about the anatomy and physiology of the neuromusculoskeletal system and the mechanics of multi-joint movement to provide a framework to reveal the cause-effect relationships between neuromuscular excitation patterns, muscle forces, ground reaction forces, and motions of the body. The models typically include detailed descriptions of musculoskeletal geometry and equations that describe the activation and force production of muscles and the multijoint dynamics of the body.

The authors identify 4 stages in the development of a muscle-driven simulation:

1. Create a model (muskuloskeletal, etc.)
2. Find a set of muscle excitations that generate the desired movement
3. Test the model by comparing results to experimental data
4. Analyze the simulation to answer research questions

The author present a somewhat simple musculoskeletal model where limbs are connected by typical joints (hinge, universal, ball-in-socket). They also present a simple but interesting muscle model that establish a relationship between:

• muscle excitation,
• muscle activations,
• muscle forces,
• musculotendon lengths and shortening velocities,
• joint torques,
• generalized coordinates and their derivatives.

They build a simplified 23 DOF character model and use experimental data to determine the physical properties of the different limbs. They model the foot-ground interactions using a series of spring-damper units distributed under the sole of the foot. The model includes 54 musculotendon actuators, the location of each being based on anatomical landmarks.

The authors use dynamic optimization with a heuristic goal. The goal is to minimize metabolic energy expenditure per unit distance traveled, penalty terms are added to discourage hyperextended joints.

Three major challenges of future work are highlighted:

• Modeling challenges
  • Simulations often sensitive to data accuracy
  • More accurate musculoskeletal geometry and minematics
  • Improve equations for musculotendon dynamics
• Simulation challenges
  • Dynamic optimization incur great computational expense (hours on 5000 processors, parallel supercomputers)
  • At best, a solution obtained in a few days or a week
  • Need efficient algorithms for subject-specific simulations
  • Adaptation of robotics control technique are promising
  • Biggest limitation: exclusion of nervous system / open-loop simulation / no reflexes.
• Analysis challenges
  • Better way to identify which muscles are responsible for an observed ground reaction force
  • Sensitivity studies to evaluate how an analysis is sensitive to the model parameters

Comments

• I should try to understand the various concepts used in the muscle model.
• What exactly is meant by dynamic optimization?
• The paper contains many references to different sources of experimental data.
• A large part of the paper is devoted to exploring how a muscle-based simulation can be used to analyze gait information that would otherwise be hard to obtain.
• A long list of contributions due to progress in muscle-based simulation is given.

-- PhilippeBeaudoin - 13 Feb 2009