Neural Network Motion Control with Backlash Threshold Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning techniques for motion control, particularly in physics-based animation and robot control, produce unnatural jittery movements and fail to account for real-world constraints, making them unsuitable for controlling physical robots, especially in conditions sensitive areas like walking and balancing.
Innovation Solution
The proposed solution involves training Neural Networks (NNs) to accept sensor readings as inputs and output either first or second derivatives of servo control commands, with integration and backlash threshold comparison to simulate real-world constraints and ensure smooth, life-like motions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning techniques are used for motion control, then automation is improved, but movement smoothness deteriorates due to jittery movements
Solution Approach 1:
The patent introduces an integrator as an intermediary component between the neural network output and the servo control system. The integrator accumulates the neural network's derivative outputs over time to reconstruct smooth position commands, effectively filtering out high-frequency jitter while preserving the automated motion control functionality.
Solution Approach 2:
The patent replaces traditional mechanical smoothing filters or PID controllers with a neural network-based derivative controller followed by integration. This substitution maintains automation while using mathematical integration to eliminate jitter, transforming the approach from mechanical/empirical control to intelligent mathematical control.
2Reliability
If real-world constraints are incorporated into the model, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent incorporates real-world constraints by modifying the training parameters and objective functions of the neural network. Instead of changing the physical hardware, the approach uses parameter adjustments in the virtual model (backlash thresholds, friction coefficients, stiffness values) to simulate real-world behavior, maintaining reliability while avoiding physical complexity.
Solution Approach 2:
The patent creates a virtual copy of the physical robot with embedded real-world constraints (backlash, friction, stiffness). This virtual model is trained separately and then applied to control the physical robot, allowing reliability to be improved through realistic simulation without directly complicating the physical device.
3Stability of the object's composition
If integration and backlash threshold comparison are implemented, then movement smoothness is improved, but computational requirements increase
Solution Approach 1:
The patent implements integration and backlash threshold comparison selectively rather than continuously at full precision. By applying these operations only when necessary (e.g., when neural network output changes significantly or thresholds are approached), the system achieves smooth movement while reducing unnecessary computational overhead and energy consumption.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables the generation of smooth, realistic movements while preserving balance and accounting for external disturbances, making the system suitable for controlling physical robots in complex environments.
Implementation Method 1
a machine learning algorithm that includes a sensor value and an integrated output value, wherein the integrated output value is compared to a backlash threshold
Data Source
AI summary
A method, system and computer program product for training a control input system involve taking an integral of an output value from a Motion Decision Neural Network for one or more movable joints to generate an integrated output value and comparing the integrated output value to a backlash threshold. A subsequent output value is generated using a machine learning algorithm that includes a sensor value and a previous joint position if the integrated output value does not at least meet the threshold. A position of the one or more movable joints is simulated based on an integral of the subsequent output value; and the Motion Decision Neural Network is trained with the machine learning algorithm based upon at least a result of the simulation of the position of the one or more movable joints.