Supervisor Network Selection for Virtual Agent Granular Motion Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training granular motion networks for every possible action that a virtual agent can perform is resource-intensive, requiring significant computing resources.
Innovation Solution
A supervisor network determines confidence scores for available granular motions and selects a subset based on these scores to advance the virtual agent towards action completion, reducing the need to train each granular motion network for every possible action.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If granular motion networks are trained for every possible action that a virtual agent can perform, then the virtual agent can execute any action with high reliability, but the computing resources required for training become excessively large
Solution Approach 1:
The patent segments the action execution system into two parts: a supervisor network that selects actions and granular motion networks that execute specific motions. Instead of training granular motion networks for every possible action, the supervisor network learns to select appropriate actions from a smaller set, dividing the computational burden and reducing training requirements while maintaining reliability
Solution Approach 2:
The supervisor network acts as an intermediary between the high-level action goal and the low-level granular motion execution. It selects which granular motion networks to activate based on the current state and desired action, reducing the need to train every granular motion network for every possible action while maintaining system reliability
Data Source
AI summary
Various implementations disclosed herein include devices, systems, and methods for granular motion control for a virtual agent. In various implementations, a device includes a non-transitory memory and one or more processors coupled with the non-transitory memory. In some implementations, a method includes obtaining an action for a virtual agent. In some implementations, the action is associated with a plurality of time frames. In some implementations, the method includes, for a first time frame of the plurality of time frames, determining respective confidence scores for a plurality of granular motions that advance the virtual agent towards completion of the action. In some implementations, the method includes selecting a subset of the plurality of granular motions based on the respective confidence scores.