Dynamic Narrow AI Agent Allocation for ADAS Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classical advanced driver assistance systems (ADAS) are rigid and operate under strict compute resource limitations, resulting in minimal acceptable quality, failing to provide more accurate solutions.
Innovation Solution
The implementation of a method that dynamically selects a subset of narrow AI agents based on time intervals and context to process sensed information, using iterative expansion and merge operations to generate multidimensional signatures for robust object detection and driving feature calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If classical ADAS systems use rigid implementation with fixed compute resource allocation, then system stability is maintained, but detection accuracy and adaptability are limited to minimal acceptable quality
Solution Approach 1:
The patent implements dynamic compute resource allocation where the system transitions from rigid fixed allocation to adaptive allocation based on real-time driving context. The compute resources are dynamically adjusted by selecting different subsets of narrow AI agents according to detected driving features, environmental conditions, and operational requirements, thereby improving detection accuracy without sacrificing system stability
Solution Approach 2:
The system changes operational parameters by varying the subset of active narrow AI agents based on driving context. Different driving scenarios (e.g., urban vs. highway, day vs. night) trigger different configurations of AI agents, allowing the system to optimize detection accuracy for each specific parameter set while managing compute resources efficiently
2Measurement precision
If more compute resources are allocated to improve detection accuracy, then driving feature detection quality improves, but resource consumption and system complexity increase
Solution Approach 1:
The patent applies local quality by activating specific subsets of narrow AI agents tailored to particular driving scenarios rather than running all agents continuously. For example, urban driving with pedestrians triggers activation of pedestrian detection specialists, while highway driving activates different agents. This localized activation improves detection quality for relevant features while conserving compute resources
Solution Approach 2:
The system implements partial action by selecting only the necessary subset of AI agents required for the current driving context rather than deploying the full suite of agents. This partial deployment achieves sufficient detection accuracy for each scenario while significantly reducing overall compute resource consumption and energy usage
3Adaptability or versatility
If a fixed subset of narrow AI agents is used, then system simplicity is maintained, but adaptability to different driving contexts and scenarios is reduced
Solution Approach 1:
The patent applies preliminary action by pre-configuring multiple narrow AI agents during system initialization, each specialized for specific driving features or scenarios. The system prepares various agent subsets in advance for different driving contexts (e.g., pedestrian detection, lane recognition, object tracking), enabling rapid adaptation to changing conditions without real-time training or complex decision-making
Solution Approach 2:
The system segments the overall AI processing task into multiple specialized narrow AI agents, each handling specific driving features. This segmentation allows the system to adapt to different contexts by selectively activating relevant segments (agents) rather than running a monolithic system, thereby improving versatility while keeping the selection mechanism manageable through modular architecture
Data Source
AI summary
A method for determining driving related features of a vehicle, the method may include: repeating, during each time interval out of multiple time intervals of a period; obtaining sensed information during at least a part of the time interval; receiving context information determined before the time interval; selecting, based on the context information, a selected sub-set of narrow artificial intelligence (AI) agents or calculating one or more driving related features. The calculating may include applying the selected sub-set of narrow AI agents on the sensed information; and determining, based on the sensed information, context information to used following the time interval.