Eurographics/SIGGRAPH Symposium on Computer Animation (2003) D. Breen, M. Lin (Editors) Flexible Automatic Motion Blending with Registration Curves Lucas Kovar 1 and Michael Gleicher 1 1 University of Wisconsin, Madison Abstract Many motion editing algorithms, including transitioning and multitarget interpolation, can be represented as instances of a more general operation called motion blending. We introduce a novel data structure called a regis- tration curve that expands the class of motions that can be successfully blended without manual input. Registration curves achieve this by automatically determining relationships involving the timing, local coordinate frame, and constraints of the input motions. We show how registration curves improve upon existing automatic blending methods and demonstrate their use in common blending operations. Categories and Subject Descriptors (according to ACM CCS): I.3.7 [Computer Graphics]: Animation 1. Introduction Motion capture has been a successful technique for creating realistic animations of humans, in large part because of the availability of motion editing methods. Since capture is fun- damentally limited to producing a finite set of clips, by itself it offers little flexibility. However, through motion editing these clips can be adapted to meet the demands of a partic- ular situation. Some of the most successful editing methods are based on the idea of motion blending, which produces new motions (blends) by combining multiple clips accord- ing to time-varying weights. Motion blending has several applications. For example, blending can be used to create seamless transitions between motions, allowing one to build lengthy, complicated motions out of simpler actions. An- other application is interpolation, or creating motions “in- between” the initial set to produce a parameterized space of motions. Blending operations such as these have significant practical importance and have proven useful in commercial applications like video games 13, 21 . While blending is a powerful tool, it will not produce re- alistic results unless the input motions are chosen with some care. The range of motions that can be successfully blended depends heavily on how much information about the mo- tions is given to the algorithm. If nothing is known about the input other than the raw motion parameters (e.g., the root po- sition and joint angles at each frame), then existing blending algorithms are reliable only if the motions are quite similar. Our goal is to expand the range of motions that can be successfully blended without requiring manual intervention. Toward this end we introduce registration curves. A registra- tion curve is an automatically constructed data structure that encapsulates relationships involving the timing, local coor- dinate frame, and constraint states of an arbitrary number of input motions. These relationships are used to improve the quality of blended motion and allow blends that were previ- ously beyond the reach of automatic methods. In the remainder of this paper we describe how to con- struct and use registration curves. The next section starts with an overview of registration curves. Section 3 reviews related work. Section 4 explains how to construct registra- tion curves, and Section 5 presents a blending algorithm that exploits the information contained in registration curves. Fi- nally, Section 6 demonstrates our technique in common ap- plications and Section 7 concludes with a discussion of the advantages and limitations of registration curves. 2. Overview of Registration Curves We represent motions in the standard skeletal format. For- mally, a motion is defined as a continuous function M( f )= (p R ( f ), q 1...n ( f )), where p R is the global position of the root and q i is the orientation of the i th joint in its parent’s coordi- nate system. Each parameter vector M( f ) is called a frame and f is the frame index. Motions are assumed to be an- notated with constraint information, e.g., that the left heel c  The Eurographics Association 2003.