Analyzing locomotion synthesis with feature-based motion graphs.

Area of Science:

• Computer Graphics
• Robotics
• Artificial Intelligence

Background:

• Traditional motion graphs for locomotion synthesis often require postprocessing to remove artifacts like foot skating.
• Computational requirements for constructing and searching traditional motion graphs can be substantial, especially in complex environments with obstacles.

Purpose of the Study:

• To introduce feature-based motion graphs (FBMGs) as a novel approach for realistic locomotion synthesis among obstacles.
• To improve search query performance, eliminate the need for postprocessing, and reduce computational demands compared to conventional motion graphs.

Main Methods:

• Developing a feature-based approach for selecting transitions, significantly reducing graph construction time and enhancing search performance.
• Implementing a fast channel search method that confines motion graph searches to clear paths among obstacles, avoiding costly collision detection.
• Introducing a motion deformation model using Inverse Kinematics applied to transitions, improving reachability and search efficiency within a user-defined cost threshold.

Main Results:

• Feature-based transition selection led to reduced graph construction time and superior search performance.
• The fast channel search method provided quicker and more effective results by ensuring clearance among obstacles.
• Motion deformation improved the overall reachability of the feature-based motion graph, further decreasing search time.

Conclusions:

• Feature-based motion graphs offer a more efficient and effective solution for realistic locomotion synthesis in cluttered environments.
• The proposed methods significantly outperform traditional motion graph techniques in terms of speed, search quality, and artifact removal.