Vehicle Motion Bias Estimation Using a Koopman Operator
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle motion control systems face biases between input commands and actual vehicle operations, leading to discrepancies such as under- or oversteering, due to mechanical variances and calibration issues, which existing solutions like Luenberger Observers and Kalman Filters often address with oscillatory behavior and sensitivity to tuning.
Innovation Solution
A vehicle system equipped with an advanced driver assistance system (ADAS) that employs a bias estimator circuit using the Koopman operator to estimate and correct biases by recursively solving nonlinear least squares equations, introducing a bias term and applying a forgetting factor to adjust for previous data points, thereby reducing sensitivity to tuning and oscillations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Luenberger Observers or Kalman Filters are used to estimate and correct steering biases, then bias correction is achieved, but the system exhibits oscillatory behavior and high sensitivity to tuning parameters
Solution Approach 1:
The patent replaces traditional mechanical control systems (Luenberger Observers and Kalman Filters) with a data-driven machine learning model (neural network) to estimate and correct steering biases. This substitution eliminates the oscillatory behavior and tuning sensitivity inherent in traditional approaches while maintaining accurate bias correction through learned patterns from training data
2Manufacturing precision
If traditional bias correction methods are applied, then steering accuracy is improved, but the system requires extensive tuning and calibration
Solution Approach 1:
The patent performs bias correction in advance by training a machine learning model on comprehensive datasets that capture various operating conditions and bias scenarios. This preliminary training phase embeds the correction logic into the model structure, eliminating the need for extensive real-time tuning and calibration during actual vehicle operation
Solution Approach 2:
The machine learning model achieves self-calibration by learning from training data that includes various bias conditions. The model automatically adapts to different vehicle configurations and operating conditions without requiring manual tuning, making the system self-sufficient and eliminating complex calibration procedures
Data Source
AI summary
A vehicle comprises: an advanced driver assistance system (ADAS) generating a command for an aspect of vehicle motion; an actuator for controlling the aspect of the vehicle motion, the actuator connected to the ADAS; and a bias estimator circuit that estimates a bias between the command for the aspect of vehicle motion and an effective vehicle motion aspect, the bias estimator circuit estimating the bias using an operator that predicts a next timestep of nonlinear functions of vehicle state, wherein the command for the aspect of vehicle motion is corrected using the bias estimated by the bias estimator circuit to provide a corrected command for the aspect of vehicle motion for use by the actuator.