Human-Like Vehicle Control Prediction for Adaptive Autonomous Driving

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current autonomous driving technologies fail to replicate human-like vehicle control behaviors, particularly in adapting to real-time situations and passenger preferences, leading to inadequate route planning and vehicle control.

Innovation Solution

A system and method for generating a human-like vehicle control model using recorded human driving data and machine learning, which considers vehicle kinematic models, real-time data, and self-aware capability parameters to produce adaptive and personalized control signals for autonomous vehicles.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional generic vehicle kinematic models and feedback controllers are used, then the control system is simple and easy to implement, but the system cannot replicate human-like driving behaviors or adapt to real-time situations and passenger preferences

Engineering Contradiction:
Improveadaptability to real-time situations and passenger preferencesVSAvoidcontrol system complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The control system is segmented into multiple specialized modules: a human-like control model generator that processes recorded human driving data, a machine learning module that creates adaptive control models, and a vehicle control module that executes personalized control signals. This segmentation allows each module to specialize in specific functions, achieving human-like adaptability while managing overall system complexity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary action by recording and analyzing human driving data beforehand to train machine learning models. These pre-trained models capture human-like driving behaviors and preferences, enabling the system to adapt to real-time situations without requiring complex real-time processing of raw data, thus balancing adaptability with computational efficiency.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If machine learning models trained on recorded human driving data are used, then human-like driving behaviors are replicated, but the training and model generation process becomes time-consuming and computationally intensive

Engineering Contradiction:
Improveaccuracy of human-like driving behavior replicationVSAvoidmodel training time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary action by recording human driving data and training machine learning models in advance, before actual autonomous driving operations. This offline training phase captures human-like driving behaviors and preferences, creating pre-trained models that can be rapidly deployed. During real-time operation, the system only needs to execute the pre-trained models, significantly reducing computational time and enabling high reliability without excessive training delays.

Inventive Principle:
Principle #10Preliminary action

3Ease of operation

If personalized control signals are generated for each passenger preference, then passenger comfort and safety are enhanced, but the processing required to analyze and respond to individual preferences increases system complexity

Engineering Contradiction:
Improvepassenger comfort and safetyVSAvoidpreference analysis and control signal generation complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The system creates simplified copies or representations of passenger preferences through machine learning models that capture essential driving behavior patterns. Instead of directly processing complex, raw preference data during real-time operation, the system uses pre-trained models that replicate human-like control decisions. This copying approach enables personalized comfort and safety control while reducing real-time processing complexity, as the models encode preference information in compact, efficient representations.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3727978B1Method and system for ensemble vehicle control prediction in autonomous driving vehicles
Publication Date: 2023.11.29 PLUSAI INC
  • EP3727978B1 patent drawingFigure 1
  • EP3727978B1 patent drawingFigure 2
  • EP3727978B1 patent drawingFigure 3

AI summary

The present teaching relates to method, system, medium, and implementation of human- like vehicle control for an autonomous vehicle. Recorded human driving data are first received, which include vehicle state data, vehicle control data, and environment data. For each piece of recorded human driving data, a vehicle kinematic model based vehicle control signal is generated in accordance with a vehicle kinematic model based on a corresponding vehicle state and vehicle control data of the piece of recorded human driving data. A human-like vehicle control model is obtained, via machine learning, based on the recorded human driving data as well as the vehicle kinematic model based vehicle control signal generated based on vehicle kinematic model. Such derived human-like vehicle control model is to be used to generate a human-like vehicle control signal with respect to a target motion of an autonomous vehicle to achieve human-like vehicle control behavior.