Character Animation

We work on automatic generation of character animation. The performance of a motion synthesizer is characterized by the fulfillment of four requirements: modest dependence on motion capture, applicability to general motions, naturalness, and accurate positioning. While none of the existing methods meets all the four requirements, we aim at developing a synthesizer that satisfies them all. In a motion editing setting, by enforcing a style similarity constraint to the reference motion, along with foot/hand position constraints, a desired motion can be generated by solving a spacetime-constrained optimization problem. To make the style comparison [spacetime optimization, resp.] more tractable, we will base its observations [variables, resp.] on a compact representation of motion. Constraints inherently met by all natural-looking human motions can be formulated to eliminate some free variables from the representation.

Trajectory Generation of Mobile Robots with Application to Robot Soccer Game

The research objective of our lab is to build a general algorithm for the trajectory generation of mobile robots that can reach from any initial configuration to arbitrary configuration. We use maximal curvature as the restriction of the vehicle in turning radius, and the path length as the criterion to search the best solution within whole combinations of trajectories. Based on the Cubic Spiral Method proposed by Y.J.Kanayama in 1997, an improved method for joining cubic spirals with straight lines is being studied now. This novel method will amend some drawbacks of the original one and be extended to suit the mobile robot with both forward and backward motion abilities. Furthermore, a basic wall-collision and obstacle-collision avoidance function is also considered for practical use.
We are planning to use this path planning method to study the ball passing problem of robot soccer games. A ball passing problem includes not only path planning but also topics of multi-robot cooperation, control strategy and velocity planning. Robot soccer games held by RoboCup have become a prevailing activity in recent years. Research institutes and famous universities all have their own robot soccer teams to participate in international games. Our lab is working hard to establish one locally.

Collision Detection of Scaled Convex Polyhedra and its Applications

This study addresses the following collision detection problem: determine the collision status for a pair of deformable convex polyhedral objects with or without relative motion, whose allowed deformation is uniform but arbitrary scaling of vertices within given upper and lower limits. This problem poses an interesting accuracy challenge to collision detection methods since a continuum, rather than a discrete set, of scaling pairs can make the objects contact each other externally. For this problem, we present a more reliable, more efficient and more accurate numerical approach as compared to reapplying GJK for each scaled configuration, based on the decision curve that characterizes the continuum of all scaling pairs that make a pair of stationary convex polyhedral objects maintain the same contact feature or make instantaneous contact feature transition. The decision curve is a descending piecewise linear curve composed of a collection of connecting line segments. Each line segment of this curve denotes the range of scaling pairs within which a specific contact feature does not vary, while the junction of two adjacent line segments, called switching point, is the single scaling pair that the contact feature changes instantaneously. This curve partitions the rectangle of allowable scaling pairs into an exact overlap and an exact non-overlap sub-region. By one-to-one mapping of this decision curve from the plane of scaling factors to the plane of parametric estimated closest points along a selected direction, collision detection could also be performed via estimated closest points. We also illustrate the family of decision curves resulting from two contacting convex polyhedra with relative motion within a range of coherent contact maintenance/transition relation is also of use as a design tool to tolerance assembly problem, a feature most distinct from GJK and other collision detection methods.Therefore the decision curve as an economical way that describes the correlations between the correlated parameters of scaling pairs, small relative motion conditions (type, direction and amount) and contact configurations (feature) of two scaled convex polyhedral objects wheredenotes the scaling factor of object i=1,2. Moreover, We illustrate via two examples this method allows users to systematically and efficiently search a lot of design of fine motions to reorient two nominally contacting convex or non-convex polyhedral objects with assigned tolerance of scaling factors that satisfy the requirements. Also the assembly problems that could make a pair of objects mate precisely under the specific circumstance, and for the collision-avoidance problems that could define the safe region of the particular relative motions for those objects.This study could also deal with the other problems that need to deal with scaling, motion and collision detection. There are many industry applications, like robot control and path planning, manufacturing/assembly, CAD/CAM design tool, interactive 3D virtual environments of TV/PC games, physically-based simulation and animation, computer graphics and movie industry…etc.