Autonomous Driving Speed Profile Generation via Cross-Modal Neural Alignment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Autonomous driving control technologies fail to reflect individual driving habits and patterns of each driver, limiting personalized speed prediction and failing to consider real-time sensor data inputs.

Innovation Solution

An autonomous driving control apparatus and method utilizing a backbone model with encoder and decoder neural networks for cross-modal alignment, incorporating road and vehicle characteristics to generate personalized speed profiles, reducing time complexity through a smaller parameter count and enabling online learning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a backbone model with encoder and decoder neural networks is used for cross-modal alignment to generate personalized speed profiles, then the accuracy of personalized speed prediction is improved, but the device complexity and computational resources required increase

Engineering Contradiction:

Improvespeed prediction accuracyVSAvoidmodel structure complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The backbone model is divided into separate encoder and decoder neural network components. The encoder processes input data (road characteristics, vehicle characteristics, driving patterns) and the decoder generates the speed profile. This segmentation allows for specialized processing in each component while maintaining overall system accuracy for personalized speed prediction.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The encoder-decoder architecture serves multiple functions: it performs cross-modal alignment between different input data types, captures personalized driving patterns, and generates optimized speed profiles. This multi-functionality reduces the need for separate specialized models for each task.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If real-time sensor data is incorporated for online learning to generate personalized speed profiles, then the adaptability to individual driving patterns is improved, but the time complexity of calculation increases

Engineering Contradiction:

Improveadaptability to driving patternsVSAvoidcalculation time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary processing of sensor data through the encoder neural network, extracting relevant features and patterns before passing them to the decoder. This preliminary action prepares the data in advance, reducing the computational burden during real-time speed profile generation and enabling faster online learning.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional mechanical calculation methods with neural network-based processing. The encoder-decoder architecture automatically learns and adapts to driving patterns through online learning, substituting complex mathematical computations with trained neural network inferences that execute faster in real-time.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If cross-modal alignment is performed to reflect distribution of road and vehicle information, then the personalization of speed profiles is improved, but the computational resources and processing requirements increase

Engineering Contradiction:

Improvepersonalization capabilityVSAvoidcomputational energy consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The encoder neural network merges multiple input modalities (road characteristics, vehicle characteristics, driving patterns) into a unified representation. This cross-modal alignment combines different data distributions into a single processed output that the decoder uses to generate personalized speed profiles, reducing the need for separate processing pipelines and lowering overall computational energy consumption.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20250276688A1Autonomous driving control apparatus and method

Publication Date: 2025.09.04 HYUNDAI MOTOR CO LTD

AI summary

An autonomous driving control apparatus includes a memory and at least one processor. The at least one processor identifies first input information including at least one of a characteristic of a road, event information of the road, or any combination thereof and second input information including a characteristic of the vehicle. The at least one processor further performs cross-modal alignment of a backbone model for outputting a speed profile. The at least one processor also applies an output obtained by applying the first input information to a first encoder neural network, an output obtained by applying the second input information to a feature extraction neural network of the backbone model, and a target speed to a second decoder neural network. The at least one processor further obtains the speed profile from the second decoder neural network to perform main learning of the backbone model.