Method and system for planning motion of virtual human

Abstract

The invention relates to a method and a system for planning the motion of virtual human. The method comprises the following steps of: firstly, obtaining the motion data of the virtual human through motion capture equipment; secondly, establishing a reinforcement learning model, wherein the reinforcement learning model comprises a state obtained through a sampling environment and an action obtained through sampling motion data, a configured state corresponds to the disposable rewards and punishment value of the action, and the state corresponds to the accumulated rewards and punishment value of the action, and initializing the accumulated rewards and punishment value; thirdly, applying the disposable rewards and punishment value and the accumulated rewards and punishment value, wherein the subsequent state of the state corresponds to the accumulated rewards and punishment value of each action, and carrying out iteration updating on the accumulated rewards and punishment value of the state, which corresponds to the action; and fourthly, for a given state of the virtual human, selecting action from the given state of the virtual human and a set target state from actions obtained by collection according to the accumulated rewards and punishment value of the state, which corresponds to the action. In the invention, different types of action segments can be selected as sample data, and time-consuming calculation can be avoided during motion synthesis.

Description

technical field [0001] The invention designs a virtual human synthesis technology, and in particular relates to a virtual human motion planning method and system thereof. Background technique [0002] In recent years, the application of virtual human synthesis technology has become increasingly widespread. From film and television animation to advertising production, from online games to sports training, from virtual maintenance to safety rehearsal, many fields involve virtual human synthesis technology. However, as the basis of these applications, how to plan and synthesize virtual human motion with certain intelligence is a technical problem that needs to be solved. [0003] Most traditional virtual character animations are manually edited and generated by experienced animators. The animator uses 3D tool software to set the static posture of the human body at the key frame moment, and then generates the intermediate posture between the key frames. Due to the high degree...

AI Technical Summary

Problems solved by technology

However, this method of reinforcement learning motion synthesis based on linear programming has the following disadvantages: (1) This method is based on such a premise that the value function can be regarded as a linear combination of a set of basis functions. If the basis function is not selected properly, cannot converge; and when the number of basis functions increases, the optimization process is inefficient

Different types of motion cannot be used for model training since they do not have the same constraint frame

Images