Humanoid Robot Action Model for Continuous Whole-Body Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotic control systems are limited in their ability to manage the vast number of degrees of freedom in humanoid robots, resulting in rigid, imprecise, and unnatural movements, particularly in unstructured environments, due to their reliance on discrete action outputs and lack of whole-body coordination.
Innovation Solution
A bipedal action model (BAM) architecture with a decoupled dual-system design, comprising an alpha model for high-level cognitive tasks and a beta model for reactive control, processes multimodal sensory data to generate continuous control commands for the robot's degrees of freedom, enabling fluid, human-like motion and dynamic adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional discrete action outputs are used to control humanoid robot degrees of freedom, then the control system structure is simple, but the movement becomes rigid and imprecise
Solution Approach 1:
The patent replaces discrete mechanical control outputs with continuous neural network-generated action vectors. The neural network outputs continuous floating-point values for each degree of freedom, substituting the traditional discrete mechanical control paradigm with a continuous, adaptive computational approach that enables fluid and precise motion control.
Solution Approach 2:
The patent changes the control parameter representation from discrete binned values to continuous floating-point parameters. By generating continuous action vectors with precise numerical values for each degree of freedom, the system achieves higher movement precision while maintaining manageable computational complexity through efficient neural network architectures.
2Stability of the object's composition
If discrete binned actions are generated for robot control, then computational resources are conserved, but temporal consistency and smoothness of motion deteriorate
Solution Approach 1:
The patent implements continuous action generation where the neural network outputs continuous floating-point values for each degree of freedom at every time step. This continuous action stream ensures temporal consistency and smooth motion transitions, eliminating the jerky movements inherent in discrete binned actions while maintaining computational efficiency through optimized network inference.
3Adaptability or versatility
If whole-body control is implemented for humanoid robots, then coordination and adaptability improve, but system complexity increases
Solution Approach 1:
The patent segments the control problem by treating each degree of freedom independently in the action vector output. The neural network generates a separate continuous control value for each joint and actuator, allowing whole-body coordination through unified continuous control while simplifying the control architecture compared to tightly coupled whole-body control methods.
4Speed
If continuous control commands are generated for all degrees of freedom, then movement fluidity improves, but computational load increases
Solution Approach 1:
The patent changes the output parameter format from discrete categories to continuous floating-point values for each degree of freedom. This parameter transformation enables fluid motion control while the computational load is managed through efficient neural network design that generates all continuous control values in a single inference pass, optimizing the balance between motion smoothness and computational efficiency.
Data Source
AI summary
A robot comprising a sensor configured to obtain data, an alpha model configured to generate data based upon both a spoken command from a human and data from the sensor, a retrieval-augmented generation module configured to obtain additional real-time knowledge from external sources, and a beta model configured to generate output data used to control an extent of the robot based in part upon the data generated by the alpha model, the additional real-time knowledge obtained by the retrieval-augmented generation module, and the data from the sensor.