Muscle activation mapping of skeletal hand motion: an evolutionary approach.

Conference: Bournemouth University, Media School.

Abstract:

Therefore, a physics engine compatible model of anatomical muscles and an alternative control architecture is essential to create biomechanical charac- ters in games.

This thesis presents a system that generates muscle activations from captured motion by borrowing principles from biomechanics and neural con- trol. A generic physics engine compliant muscle model primitive is also de- veloped. The muscle model primitive forms the motion actuator and is an integral part of the physical model used in the simulation.

This thesis investigates a stochastic solution to create a controller that mimics the neural control system employed in the human body. The control system uses evolutionary neural networks that evolve its weights using genetic algorithms. Examples and guidance often act as templates in muscle training during all stages of human life. Similarly, the neural con- troller attempts to learn muscle coordination through input motion samples.

The thesis also explores the objective functions developed that aids in the genetic evolution of the neural network.

Character interaction with the game world is still a pre-animated behaviour in most current games. Physically-based procedural hand ani- mation is a step towards autonomous interaction of game characters with the game world. The neural controller and the muscle primitive developed are used to animate a dynamic model of a human hand within a real-time physics engine environment.

https://eprints.bournemouth.ac.uk/20980/

Source: Manual

Preferred by: Arun Somasekharan

Muscle activation mapping of skeletal hand motion: an evolutionary approach.

Authors: Somasekharan, A.

Conference: Bournemouth University

Pages: ?-? (225)

Abstract:

Creating controlled dynamic character animation consists of mathe- matical modelling of muscles and solving the activation dynamics that form the key to coordination. But biomechanical simulation and control is com- putationally expensive involving complex di erential equations and is not suitable for real-time platforms like games. Performing such computations at every time-step reduces frame rate. Modern games use generic soft- ware packages called physics engines to perform a wide variety of in-game physical e ects. The physics engines are optimized for gaming platforms. Therefore, a physics engine compatible model of anatomical muscles and an alternative control architecture is essential to create biomechanical charac- ters in games. This thesis presents a system that generates muscle activations from captured motion by borrowing principles from biomechanics and neural con- trol. A generic physics engine compliant muscle model primitive is also de- veloped. The muscle model primitive forms the motion actuator and is an integral part of the physical model used in the simulation. This thesis investigates a stochastic solution to create a controller that mimics the neural control system employed in the human body. The control system uses evolutionary neural networks that evolve its weights using genetic algorithms. Examples and guidance often act as templates in muscle training during all stages of human life. Similarly, the neural con- troller attempts to learn muscle coordination through input motion samples. The thesis also explores the objective functions developed that aids in the genetic evolution of the neural network. Character interaction with the game world is still a pre-animated behaviour in most current games. Physically-based procedural hand ani- mation is a step towards autonomous interaction of game characters with the game world. The neural controller and the muscle primitive developed are used to animate a dynamic model of a human hand within a real-time physics engine environment.

https://eprints.bournemouth.ac.uk/20980/

Source: BURO EPrints