Humanoid Bipedal 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 dynamic environments, due to their reliance on discrete action outputs and lack of whole-body coordination.
Innovation Solution
A bipedal action model (BAM) that integrates a high-level cognitive alpha model and a low-level reactive beta model, processing multimodal sensory data and natural language instructions to generate continuous control commands for the robot's degrees of freedom, enabling fluid, human-like motion through a decoupled dual-system design and layered training data structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional discrete action outputs are used to control humanoid robot, then the control system is simpler to implement, but the movement becomes jerky, imprecise, and unnatural
Solution Approach 1:
The patent replaces the conventional discrete mechanical control system with a neural network-based continuous control system. The BAM uses neural networks to generate continuous action outputs that directly specify control parameters for each degree of freedom, eliminating the need for discrete action binning and enabling smooth, precise, and natural robot movements.
Solution Approach 2:
The patent changes the control output from discrete binned values to continuous floating-point parameters. Instead of selecting from predefined action categories, the system outputs continuous control parameters that can take any value within a range, enabling fine-grained control and eliminating the jerky movements caused by discrete action transitions.
2Power
If discrete binned actions are generated, then computational processing is reduced, but temporal consistency is lost and compounding errors occur
Solution Approach 1:
The patent implements continuous action generation where the neural network outputs continuous control commands for each time step. This continuous approach maintains temporal consistency across the action sequence, preventing the compounding errors that occur when discrete actions are executed sequentially. The system generates a smooth continuum of control outputs rather than discrete snapshots.
3Adaptability or versatility
If whole-body control is implemented to coordinate entire robot body, then the system can perform dynamic balance and complex tasks, but the device complexity increases significantly
Solution Approach 1:
The patent segments the control problem by treating each degree of freedom independently. The neural network outputs a separate control parameter for each joint and actuator, allowing whole-body coordination without requiring complex inter-joint coupling models. This segmentation simplifies the control architecture while enabling comprehensive whole-body control.
Solution Approach 2:
The patent creates a universal control framework that handles all degrees of freedom through a single neural network architecture. The BAM can control any combination of joints, limbs, and actuators using the same continuous output mechanism, providing multi-functional capability without requiring separate control systems for different body parts.
4Manufacturing precision
If continuous control commands are generated for all degrees of freedom, then movement smoothness is improved, but computational resources and model complexity increase
Solution Approach 1:
The patent changes the output parameters from discrete action categories to continuous floating-point values for each degree of freedom. This parameter transformation enables smooth movement generation while the neural network architecture efficiently handles the increased computational requirements through parallel processing of continuous values.
Data Source
AI summary
The present disclosure provides a humanoid robot comprising a torso having an alpha model deployed on a first GPU, and wherein said alpha model includes a first number of parameters and is configured to receive a natural language command from a human and generate processed data, a beta model deployed on a second GPU, and wherein said beta model includes a second number of parameters and is configured to receive the processed data from the alpha model and provide output data used to control an extent of the left wrist, and wherein the first number of parameters is larger than the second number of parameters, and a unified training framework is used to jointly train the alpha model and the beta model.