SVM Classifier for Biped Robot Fall Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring the balanced state of humanoid robots are limited, as they can only perform accurately when the robot is standing stably and require extensive data collection, which is difficult and damaging to obtain, especially during falls.
Innovation Solution
A method using a support vector machine (SVM) classifier trained with state data from stable and falling robot states, incorporating angular velocity, linear acceleration, and center of mass projection distance, allowing for real-time balanced state monitoring during motion with reduced data collection and damage risk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning method with 16 classifiers is used to predict humanoid falls, then prediction capability during motion is improved, but data collection complexity and robot damage risk increase significantly
Solution Approach 1:
The patent changes the parameter dimension by selecting 7 key state parameters (three-dimensional position and four-dimensional attitude) to form the input feature vector for the SVM classifier. This parameter selection strategy simplifies the data collection process while maintaining effective fall prediction capability during robot motion.
Solution Approach 2:
The patent uses support vector machine classification to create a computational model that copies and generalizes fall patterns from training data. The SVM classifier learns from labeled training samples and can predict falls in unseen situations, eliminating the need to collect extensive new data for each scenario while maintaining high prediction reliability.
2Measurement precision
If physical model-based methods are used to monitor balanced state, then stable standing monitoring is improved, but prediction accuracy deteriorates when contact points slide or during motion
Solution Approach 1:
The patent transitions from static balanced state monitoring to dynamic fall prediction by using SVM classification on time-series state data. The method processes sequential state information and can predict falls during robot motion, making the system adaptable to dynamic conditions while maintaining monitoring precision through the buffer time mechanism.
Solution Approach 2:
The patent implements a feedback mechanism by using the SVM classifier to continuously monitor robot state and predict potential falls. The system processes current state data, compares it against learned patterns from training, and provides prediction feedback that can trigger protective actions, enabling accurate monitoring during both stable standing and motion phases.
3Measurement precision
If extensive fall data is collected for training, then classifier accuracy is improved, but robot damage risk during data collection increases
Solution Approach 1:
The patent treats simulation data and synthetic training samples as disposable resources that can be generated without physical risk. By using simulation environments and synthetic data generation, the system obtains sufficient training data without repeatedly subjecting the physical robot to damaging fall conditions, thus maintaining classifier accuracy while eliminating damage risk.
Solution Approach 2:
The patent converts the harmful aspect of actual falls into a benefit by using simulation and synthetic data generation techniques. Instead of collecting data through actual damaging falls, the system generates equivalent training data computationally, transforming what would be a harmful data collection process into a safe and efficient one while maintaining training effectiveness.
Data Source
AI summary
The present invention provides a method for monitoring a balanced state of a humanoid robot, comprising: acquiring state data of the robot falling in different directions and being stable, forming a support vector machine (SVM) training data set and obtaining, by training, an initial SVM classifier; inputting the state data of the robot to the trained SVM classifier, so that the SVM classifier outputs a classification result; taking statistics on a proportion of cycles judged to have an impending fall in the total number of control cycles within a judgment buffer time after the SVM classifier outputs the classification result, and finally determining a monitoring result of the balanced state of the robot according to the proportion and finally extracting state data of misjudged cycles within the buffer time, adding the state data to the current training data set and updating the SVM classifier, eventually enabling the classifier to achieve the effects of matching motion capabilities of the robot and monitoring the balanced state.