Optimized Computer-Generated Motions for Animation

Goldsmith, 1994

Category: Computer Graphics

Overall Rating

3.3/5 (23/35 pts)

This paper offers a unique, actionable path not through its general optimization framework, but via a specific empirical finding: minimizing the volume-integrated covariant acceleration of an articulated body reportedly produces 'anticipatory' and fluid motion.
This specific objective function and its qualitative outcome appear distinct from standard animation or robotics metrics and could be a niche 'gem' for generating specific aesthetic motion qualities in modern systems, leveraging current computational power.
The critical assessment correctly points out the severe limitations of the general optimization approach and its computational cost on older hardware, and that many aspects have been superseded.
However, the specific objective function explored for generating perceptually fluid motion remains an interesting, potentially underexplored niche application for domains like HRI.

This thesis explores using constrained optimization to generate animated motion for articulated figures by minimizing various objective functions, notably the integral of the square of the (covariant) acceleration of points on the body.
The key finding highlighted is that minimizing the integral of the covariant acceleration over the volume of the body... produces 'anticipatory' and 'fluid-appearing' motions.
A potentially high-impact, unconventional research direction lies in applying this specific objective function... within the context of modern robot motion planning and human-robot interaction (HRI).
This is unconventional because it shifts the focus from purely functional robot metrics... to a geometrically and physically informed aesthetic metric that explicitly promotes a specific, perceptually desirable quality of motion ('anticipatory,' 'fluid').

The core paradigm of this work—offline optimization of simple physical/geometric objective functions for keyframe interpolation—has largely decayed in relevance for many modern animation tasks.
The paper itself acknowledges significant practical limitations that likely contributed to its lack of lasting impact: Computational Cost, Numerical Optimization Difficulties, Constraint Handling, Limited Generality of Objectives.
Its paradigm of offline keyframe interpolation has been superseded by faster, more robust, and more controllable modern animation methods.
Applying this paper's specific optimization-based keyframe interpolation approach to unrelated fields like general AI, machine learning for motor control... would be a significant pitfall.

Watch