Road Object Understanding Pipeline for Unknown AV Obstacles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles (AVs) face challenges in accurately classifying and understanding road objects, particularly those that are unknown or misclassified, which can lead to inefficient navigation and increased risk of getting stuck or requiring physical retrieval.
Innovation Solution
A road object understanding model is introduced as a sub-model within the AV's perception and understanding pipeline, utilizing a shared backbone and multiple heads to classify road objects and extract attributes, improving precision and recall metrics, and reducing computational complexity by processing cropped sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a main understanding model classifies tracked objects into road user classifications and unknown object classification, then the AV can identify basic object categories, but the precision and recall for unknown road objects remain insufficient leading to navigation inefficiency and potential getting stuck
Solution Approach 1:
The patent divides the object classification system into two distinct models: a main understanding model that handles common road users, and a specialized road object understanding model that handles unknown road objects. This segmentation allows each model to be optimized for its specific task, improving overall classification precision without requiring a single overly complex model to handle all object types equally well.
Solution Approach 2:
The patent introduces a classification module as an intermediary that determines whether a tracked object should be processed by the main understanding model or the road object understanding model. This intermediary routing mechanism ensures that unknown road objects are directed to the appropriate specialized model, improving recall and precision for these objects without burdening the main model with excessive complexity.
2Measurement precision
If the AV processes all sensor data through comprehensive classification models, then classification accuracy improves, but computational complexity increases leading to longer processing time
Solution Approach 1:
The patent applies partial action by having the road object understanding model process only cropped sensor data corresponding to tracked objects that the main model classifies as unknown. This selective processing approach maintains high classification accuracy for unknown objects while significantly reducing the computational load and processing time compared to analyzing all sensor data at full resolution through both models.
Solution Approach 2:
The patent segments the data processing workflow into two stages: initial classification by the main model, then selective detailed analysis by the road object model only for unknown objects. This segmentation of the processing pipeline ensures that computationally intensive operations are performed only when necessary, optimizing the balance between accuracy and processing time.
3Reliability
If the AV uses detailed road object classification and attribute extraction, then navigation decision quality improves, but the device complexity and computational requirements increase
Solution Approach 1:
The patent segments the understanding system into specialized components: a main understanding model for common objects, a dedicated road object understanding model for unknown objects, and a separate classification module for attribute extraction. This segmentation allows each component to be optimized for its specific function, improving navigation reliability through specialized processing while managing system complexity through modular design.
Solution Approach 2:
The road object understanding model automatically activates only when the main model identifies unknown objects, and the classification module self-regulates by processing only the necessary cropped data. This self-service mechanism ensures detailed classification and attribute extraction occur only when needed for navigation decisions, improving reliability without requiring the system to permanently maintain high computational complexity.
Data Source
AI summary
Autonomous vehicles utilize perception and understanding of road users and road objects to predict behaviors of the road users and road objects, and to plan a trajectory for the vehicle. Improved perception and understanding of the AV's surroundings can improve the AV's behavior around drivable objects, non-drivable road objects, construction zones, and temporary road closures. Improved perception and understanding of the AV's surroundings can also reduce the chances of the AV getting stuck and the need to be retrieved physically. To offer additional understanding capabilities, an additional understanding model is added to the perception and understanding pipeline to improve classification of road objects and extraction of attributes of the road objects. The implementation of the understanding model itself and placement of the model within the pipeline balance recall and precision performance metrics and computational complexity.