Hidden Markov Driving Variable Prediction for Driver Torque

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for predicting driving variables in driver-vehicle systems, such as the steering wheel angle, are limited and do not effectively calculate other important variables like driver torque or pressure on the accelerator and brake pedals, which are nonlinear and dependent on driver behavior, requiring more accurate and adaptive solutions.

Innovation Solution

A computer-implemented method using a Hidden Markov Model (HMM) that predicts driving variables at future time steps by inputting current feature values, calculating probabilities based on state transition and emission distributions, and incorporating adaptive capabilities to handle various drivers and constraints like smoothness, allowing for real-time calculations with high accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a Hidden Markov Model is used to predict driving variables, then the accuracy of prediction for variables like driver torque and pedal pressure is improved, but the complexity of the system increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the prediction task by separating the Hidden Markov Model (which handles temporal dependencies and state transitions) from the observation features (which capture current driving context). This segmentation allows each component to specialize, improving prediction accuracy while keeping the overall architecture manageable through modular design

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The Hidden Markov Model serves multiple functions simultaneously: it models temporal dynamics, captures driver behavior patterns, and handles multiple driving variables (steering torque, accelerator pressure, brake pressure) within a unified framework. This multi-functionality improves prediction accuracy across different variables without requiring separate models for each

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If the system predicts multiple driving variables including driver torque and pedal pressure, then the versatility of the system is improved, but the computational requirements and complexity increase

Engineering Contradiction:
Improvesystem versatilityVSAvoidcomputational complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

A single Hidden Markov Model framework is designed to predict multiple driving variables (steering wheel torque, accelerator pedal pressure, brake pedal pressure) simultaneously. The model uses a unified set of observation features and state transition mechanisms that work across all variables, achieving high versatility without requiring separate specialized models for each variable

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system adapts to different driving variables by changing the observation space and emission probability parameters of the HMM rather than changing the fundamental model structure. This allows the same architectural framework to handle multiple variables with different characteristics, improving versatility while controlling computational complexity through parameter adaptation rather than structural expansion

Inventive Principle:
Principle #35Parameter changes

3Speed

If real-time prediction is implemented, then the responsiveness of the system is improved, but the computational load increases

Engineering Contradiction:
Improveprediction speedVSAvoidcomputational load
Core Design Contradiction:
SpeedVSUse of energy by moving object

Solution Approach 1:

The Hidden Markov Model parameters (transition probabilities, emission distributions) are pre-computed and stored during an offline training phase using historical driving data. During real-time operation, the system only needs to perform efficient probability calculations using these pre-computed parameters, rather than performing complex learning computations. This preliminary action enables fast real-time predictions while keeping online computational load manageable

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4280123A1Driving variable calculation method and system, input features combination selection method, number of states determination method, program, storage medium and system
Publication Date: 2023.11.22 TOYOTA JIDOSHA KK
  • EP4280123A1 patent drawingFigure 1
  • EP4280123A1 patent drawingFigure 2
  • EP4280123A1 patent drawingFigure 3A

AI summary

A driving variable calculation method for calculating a driving variable (Testd,t) related to a state of a driver-vehicle system during driving at a next time step (t+1); a hidden Markov Model (HMM) being defined by a number of states (sk), state transition probabilities and state emission distributions (bk); each state emission distribution (bk) being a probability distribution for the driver-vehicle system of being in a state (sk) and having a current combined observation vector (xt) combining current values of input features (F) and said at least one driving variable (Tdt); the method comprising steps of: S10) inputting current values of input features (F); S20) based on said vector, calculating said at least one driving variable (Td) at said next time step (t+1) based on the hidden Markov Model (HMM). The invention further encompasses a system, a program, a data storage medium, adapted to perform the driving variable calculation method.