Autonomous Vehicle Behavior Control via Deep Learning Pattern Prediction

Solution Overview

Problem

Autonomous vehicles experience sudden behavior changes due to temporal gaps in control value transmission, leading to rapid acceleration, deceleration, or steering, causing anxiety for occupants, especially in low-speed congestion or high-speed driving scenarios.

Innovation Solution

A behavior control device and method that performs deep learning of vehicle behavior patterns using combined sensor data, including steering wheel angle and path information, to predict and control speed or steering based on the driving environment, reducing the need for immediate fusion data and stabilizing vehicle behavior.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the controller generates control values at a high transmission period (100 ms) to reduce computational load, then processing efficiency is improved, but the control values do not reflect real-time vehicle behavior, causing instability and sudden movements

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidvehicle behavior stability

Core Design Contradiction:

ProductivityVSStability of the object's composition

Solution Approach 1:

The learning device performs deep learning of behavior patterns in advance and stores them for later use. When fusion data is received, the controller applies these pre-learned patterns to generate control values at high frequency (10 ms), eliminating the need for real-time complex processing while maintaining behavioral accuracy and stability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The learning device acts as an intermediary between the fusion device and controller. It processes fusion data at lower frequency, learns behavior patterns, and provides the controller with preprocessed information that enables stable high-frequency control without requiring real-time deep learning computation.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Speed

If the controller transmits control values at a high frequency (10 ms) to ensure responsive control, then control responsiveness is improved, but the control values may be based on outdated fusion data, causing sudden behavior changes and occupant anxiety

Engineering Contradiction:

Improvecontrol responsivenessVSAvoidcontrol value accuracy

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

The learning device performs deep learning of behavior patterns in advance and stores them for later use. When fusion data is received, the controller applies these pre-learned patterns to generate control values at high frequency (10 ms), eliminating the need for real-time complex processing while maintaining behavioral accuracy and stability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The learning device continuously learns from incoming fusion data and updates behavior patterns. This feedback mechanism ensures that even when generating control values at high frequency, the controller uses the most recently learned patterns that reflect current vehicle behavior, maintaining both responsiveness and accuracy.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If the system uses deep learning to accurately predict vehicle behavior, then control accuracy is improved, but the computational complexity and processing time increase significantly

Engineering Contradiction:

Improvebehavior prediction accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The learning device performs deep learning of behavior patterns in advance and stores them for later use. When fusion data is received, the controller applies these pre-learned patterns to generate control values at high frequency (10 ms), eliminating the need for real-time complex processing while maintaining behavioral accuracy and stability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system separates the computationally intensive deep learning process from the real-time control process. The learning device handles complex pattern recognition offline or at lower frequency, while the controller handles simple pattern application at high frequency, dividing the computational burden into manageable segments.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11507093B2Behavior control device and behavior control method for autonomous vehicles

Publication Date: 2022.11.22 HYUNDAI MOTOR CO LTD

AI summary

A behavior control device and a behavior control method for an autonomous vehicle are provided. The behavior control device includes a learning device configured to perform deep learning of a behavior pattern of a vehicle according to a driving environment and a controller configured to control a behavior of the autonomous vehicle based on a result of the learning of the learning device.