Vehicle Perception Network Switching for Situation-Specific ADS Sensing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The limited computational resources in vehicles pose a challenge for deploying high-performance environment perception capabilities in autonomous and semi-autonomous vehicles, as existing neural networks require significant computational power, leading to suboptimal performance in specific situations.

Innovation Solution

Implement a system with a generalized perception network and multiple specific perception networks, each trained for different traffic situations, allowing selective activation based on current conditions to enhance situational performance while minimizing computational demands.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a generalized perception network is used to handle all traffic situations, then the system can operate with a single model, but the performance in specific situations is suboptimal

Engineering Contradiction:
Improveperception performanceVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The perception system is segmented into a generalized perception network that handles all situations and multiple specific perception networks, each trained for particular traffic situations. The system divides the workload by creating specialized sub-networks (e.g., for nighttime driving, rainy conditions, highway scenarios) that are activated only when their specific conditions are detected, thereby improving reliability for each scenario without requiring a single overly complex model to handle everything perfectly

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically selects which specific perception network to activate based on the current traffic situation detected by the generalized network. This dynamic switching mechanism allows the system to adapt its complexity level - using simple generalized processing for common situations and activating more specialized networks only when needed, thus improving overall performance without permanently increasing system complexity

Inventive Principle:
Principle #15Dynamics

2Reliability

If multiple specific perception networks are deployed for different traffic situations, then situational performance is improved, but computational resource requirements increase

Engineering Contradiction:
Improvesituational performanceVSAvoidcomputational power
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The computational workload is segmented across multiple specialized networks rather than concentrating all processing in one large model. Each specific perception network is trained for a narrow set of conditions, requiring fewer computational resources individually. The system only activates the specific network matching the current situation, so computational power usage remains low for most operations while maintaining high performance when needed

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of deploying all specific perception networks simultaneously (excessive action), the system activates only the specific network relevant to the current traffic situation (partial action). This selective activation ensures that computational resources are used efficiently - full computational power is dedicated to the single most relevant specialized network rather than dividing resources across multiple networks or running them all in parallel

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4361967B1Situation specific perception capability for autonomous and semi-autonomous vehicles
Publication Date: 2026.03.11 ZENSEACT AB
  • EP4361967B1 patent drawingFigure 1~2
  • EP4361967B1 patent drawingFigure 3
  • EP4361967B1 patent drawingFigure 4

AI summary

A computer-implemented method and related aspects for providing environment perception for a vehicle having an Automated Driving System (ADS) are disclosed. The ADS comprises a generalized perception network and a plurality of specific perception networks, where each specific perception network comprises a specifically trained machine-learning algorithm that is trained with data associated with a specific traffic situation such that each specifically trained machine-learning algorithm is trained for a different traffic situation as compared to the other specifically trained machine-learning algorithms of the plurality of specific perception networks. The method comprises obtaining sensor data from one or more vehicle-mounted sensors. The obtained sensor data comprises information about a surrounding environment of the vehicle. The method further comprises selecting one or more specific perception networks for activation out of the plurality of specific perception networks based on a current traffic situation indicated in the obtained sensor data. Furthermore, the method comprises feeding at least a portion of the obtained sensor data to each of the selected one or more specific perception networks and to the generalized perception network in order to obtain an output from each of the selected one or more specific perception networks and the generalized perception network. The method further comprises fusing the obtained outputs from each of the selected one or more specific perception networks and the generalized perception network, wherein the fused output comprises one or more perception output in relation to the surrounding environment of the vehicle. Further, the method comprises transmit the fused output to a module of the ADS.