Method and system for real-time animation generation using machine learning

1. A system comprising:

one or more computer processors;

one or more computer memories;

a set of instructions incorporated into the one or more computer memories, the set of instructions configuring the one or more computer processors to perform operations comprising:

determining a dominant atom from a set of active atoms associated with a character based on weights associated with the set of active atoms, each atom of the set of active atoms including a reference to data defining a character pose within a motion library;

determining a motion controlling ability from a plurality of abilities of the character based on priorities of the abilities and a current game state produced within a game;

determining a motion fragment for the dominant atom based on pose data and future trajectory data in the motion library associated with the dominant atom;

calculating a predicted future trajectory of the character based on the motion fragment, the controlling ability, and control data from an input device;

based on a detection of a future collision or problem between the predicted future trajectory and an environment within the game, calculating a combined future trajectory based on the predicted future trajectory and an additional future trajectory;

determining a combined motion fragment by replacing the future trajectory of the motion fragment with the predicted future trajectory or combined future trajectory;

converting the combined motion fragment to a hash value using a hashing function and using a nearest neighbor search to find the closest match between the hash value and a second hash value in the motion library;

adding the atom associated with the second hash value to the set of active atoms as a target atom;

posing the character in a frame based on a weighted combination of poses associated with the set of active atoms; and

changing the weights associated with the set of active atoms based on a time function, the changing including increasing the weight of the target atom and decreasing the weight of all other atoms.

2. The system of claim 1 , wherein the predicted future trajectory and the additional future trajectory are determined using a trajectory prediction neural network trained on animation data to predict trajectories of animated characters.

3. The system of claim 1 , wherein the motion library comprises a plurality of animation clip data organized into motion fragments centered on a plurality of poses in the motion library, and with the motion fragments converted to hash values by the hashing function.

4. The system of claim 1 , wherein the additional future trajectory is determined starting from a location of the collision or problem and with a new direction to avoid the collision or problem.

5. The system of claim 1 , wherein the motion fragment includes a plurality of character poses from the motion library preceding the dominant atom and a future trajectory following the dominant atom.

6. The system of claim 1 , wherein the converting the combined future trajectory to a hash value includes using a hash function determined by a siamese hash neural network trained on motion fragments within the motion library.

7. The system of claim 2 , wherein the trajectory prediction neural network is trained on a type of motion, and the motion controlling ability is associated with the type of motion, and has a priority based on the type of motion.

8. A method comprising:

determining a dominant atom from a set of active atoms associated with a character based on weights associated with the set of active atoms, each atom of the set of active atoms including a reference to data defining a character pose within a motion library;

determining a motion controlling ability from a plurality of abilities of the character based on priorities of the abilities and a current game state produced within a game;

determining a motion fragment for the dominant atom based on pose data and future trajectory data in the motion library associated with the dominant atom;

calculating a predicted future trajectory of the character based on the motion fragment, the controlling ability, and control data from an input device;

based on a detection of a future collision or problem between the predicted future trajectory and an environment within the game, calculating a combined future trajectory based on the predicted future trajectory and an additional future trajectory;

determining a combined motion fragment by replacing the future trajectory of the motion fragment with the predicted future trajectory or combined future trajectory;

converting the combined motion fragment to a hash value using a hashing function and using a nearest neighbor search to find the closest match between the hash value and a second hash value in the motion library;

adding the atom associated with the second hash value to the set of active atoms as a target atom;

posing the character in a frame based on a weighted combination of poses associated with the set of active atoms; and

changing the weights associated with the set of active atoms based on a time function, the changing including increasing the weight of the target atom and decreasing the weight of all other atoms.

9. The method of claim 8 , wherein the predicted future trajectory and the additional future trajectory are determined using a trajectory prediction neural network trained on animation data to predict trajectories of animated characters.

10. The method of claim 8 , wherein the motion library comprises a plurality of animation clip data organized into motion fragments centered on a plurality of poses in the motion library, and with the motion fragments converted to hash values by the hashing function.

11. The method of claim 8 , wherein the additional future trajectory is determined starting from a location of the collision or problem and with a new direction to avoid the collision or problem.

12. The method of claim 8 , wherein the motion fragment includes a plurality of character poses from the motion library preceding the dominant atom and a future trajectory following the dominant atom.

13. The method of claim 8 , wherein the converting the combined future trajectory to a hash value includes using a hash function determined by a siamese hash neural network trained on motion fragments within the motion library.

14. The method of claim 9 , wherein the trajectory prediction neural network is trained on a type of motion, and the motion controlling ability is associated with the type of motion, and has a priority based on the type of motion.

15. A non-transitory machine-readable medium storing a set of instructions that, when executed by at least one processor, cause the at least one processor to perform operations comprising:

determining a dominant atom from a set of active atoms associated with a character based on weights associated with the set of active atoms, each atom of the set of active atoms including a reference to data defining a character pose within a motion library;

determining a motion controlling ability from a plurality of abilities of the character based on priorities of the abilities and a current game state produced within a game;

determining a motion fragment for the dominant atom based on pose data and future trajectory data in the motion library associated with the dominant atom;

calculating a predicted future trajectory of the character based on the motion fragment, the controlling ability, and control data from an input device;

based on a detection of a future collision or problem between the predicted future trajectory and an environment within the game, calculating a combined future trajectory based on the predicted future trajectory and an additional future trajectory;

determining a combined motion fragment by replacing the future trajectory of the motion fragment with the predicted future trajectory or combined future trajectory;

converting the combined motion fragment to a hash value using a hashing function and using a nearest neighbor search to find the closest match between the hash value and a second hash value in the motion library;

adding the atom associated with the second hash value to the set of active atoms as a target atom;

posing the character in a frame based on a weighted combination of poses associated with the set of active atoms; and

changing the weights associated with the set of active atoms based on a time function, the changing including increasing the weight of the target atom and decreasing the weight of all other atoms.

16. The non-transitory machine-readable medium of claim 15 , wherein the predicted future trajectory and the additional future trajectory are determined using a trajectory prediction neural network trained on animation data to predict trajectories of animated characters.

17. The non-transitory machine-readable medium of claim 15 , wherein the motion library comprises a plurality of animation clip data organized into motion fragments centered on a plurality of poses in the motion library, and with the motion fragments converted to hash values by the hashing function.

18. The non-transitory machine-readable medium of claim 15 , wherein the additional future trajectory is determined starting from a location of the collision or problem and with a new direction to avoid the collision or problem.

19. The non-transitory machine-readable medium of claim 15 , wherein the motion fragment includes a plurality of character poses from the motion library preceding the dominant atom and a future trajectory following the dominant atom.

20. The non-transitory machine-readable medium of claim 15 , wherein the converting the combined future trajectory to a hash value includes using a hash function determined by a siamese hash neural network trained on motion fragments within the motion library.