Trajectory optimization for full-body movements with complex contacts.

Area of Science:

• Computer Graphics
• Robotics
• Animation

Background:

• Physics-based human motion synthesis is crucial for realistic animation and robotics.
• Existing methods often require motion capture, key-poses, or periodic motion, limiting their applicability.
• Synthesizing complex, non-periodic full-body movements remains a challenge.

Purpose of the Study:

• To present a new method for full-body trajectory optimization of physics-based human motion.
• To enable synthesis of diverse movements without reliance on external data like motion capture or predefined poses.
• To achieve motion synthesis using a minimal set of simple, high-level goals.

Main Methods:

• The method optimizes full-body trajectories using physics simulation.
• Optimization is guided by simple objectives (e.g., center-of-mass height, contact points) applied over short spacetime windows.
• These windows are composed to achieve complex, long-duration motions.
• The approach does not require specifying contact locations beforehand.

Main Results:

• The method successfully synthesized a wide range of human movements, including walking, hand walking, breakdancing, flips, and crawling.
• These synthesized motions were achieved without motion capture, key-poses, or periodic motion assumptions.
• A significant portion of the demonstrated movements represent novel achievements for physics-based synthesis methods.

Conclusions:

• This work offers a powerful and flexible approach to physics-based human motion synthesis.
• The method's ability to generate diverse, complex motions from simple goals opens new possibilities for animation and virtual character control.
• It overcomes limitations of previous techniques, paving the way for more accessible and versatile motion generation.