Vehicle Environment Pattern Detection With Multi-Scale Attention

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning algorithms for autonomous driving and ADAS struggle with low flexibility in adapting to varying environmental conditions when determining spatial-temporal patterns from sensor data, leading to suboptimal performance in dynamic scenarios.

Innovation Solution

A method utilizing an attention algorithm to dynamically combine spatial-temporal data across multiple processing levels with different scales, employing a query vector independent of processing levels and key and value vectors on each level to adapt to varying conditions, enhancing the flexibility and accuracy of pattern recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a static recombination of spatial-temporal patterns is used across multiple scales, then the processing structure is simple, but the flexibility to adapt to varying environmental conditions is low

Engineering Contradiction:
Improveflexibility to adapt to varying conditionsVSAvoidcomplexity of processing structure
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies dynamics by transforming the static recombination structure into a dynamic one through the introduction of attention mechanisms. The attention weights are computed based on query, key, and value vectors that adapt to the current input data and environmental conditions, allowing the system to dynamically adjust the importance of different spatial-temporal patterns at various scales rather than using fixed static combinations.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes parameters by introducing learnable attention weights that modulate the contribution of each processing level. These weights are computed based on the interaction between query vectors and key vectors, allowing the system to adaptively adjust the parameters (attention weights) according to the specific environmental conditions and input data characteristics, thereby improving flexibility without rigid structural changes.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple processing levels with different scales are used, then the maximum resolvable velocities are increased, but the computational complexity increases

Engineering Contradiction:
Improvemaximum resolvable velocitiesVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the most relevant information from multiple processing levels by using attention mechanisms to selectively focus on important spatial-temporal patterns. Instead of processing and combining all data from all scales equally, the attention mechanism extracts and prioritizes the most significant features based on their relevance to the current query, reducing unnecessary computational overhead while maintaining high velocity resolution.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by not fully processing all processing levels with equal effort. The attention mechanism allows the system to partially process less relevant scales while focusing computational resources on the most informative scales, thereby achieving high measurement precision for velocities without the full computational cost of equally processing all levels.

Inventive Principle:
Principle #16Partial or excessive action

3Reliability

If recurrent units process data at each scaling level separately, then spatial-temporal correlations are exploited effectively, but the flexibility for dynamic adaptation is reduced

Engineering Contradiction:
Improvespatial-temporal correlation exploitationVSAvoiddynamic adaptation capability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent introduces feedback through the attention mechanism that uses the output of recurrent units at different scales to compute attention weights, which then feedback into the aggregation process. The query vectors interact with key vectors from recurrent unit outputs to determine attention weights, creating a feedback loop that allows the system to dynamically adapt while maintaining effective exploitation of spatial-temporal correlations captured by the recurrent units.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4485379B1Method for determining spatial-temporal patterns related to the environment of a vehicle
Publication Date: 2026.01.14 APTIV TECHNOLOGIES AG
  • EP4485379B1 patent drawingFigure 1
  • EP4485379B1 patent drawingFigure 2
  • EP4485379B1 patent drawingFigure 3

AI summary

A method is provided for determining patterns related to an environment of a host vehicle. Characteristics are detected by a perception system of the host vehicle within the environment of the host vehicle. At least two processing levels having different scales are defined. For each processing level, respective current input data associated with the characteristics for a current point in time is combined with respective memory data related to the characteristics for previous points in time in order to generate joint spatial-temporal data for the respective processing level. An attention algorithm is applied to the joint spatial-temporal data of all processing levels for generating an aggregated data set, and at least one pattern related to the environment is determined from the aggregated data set.