Self-Driving Car Path Graph Generation With Cached Static Attributes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for Self-Driving Cars (SDCs) face challenges in efficiently detecting objects around the vehicle, particularly in generating graph-structures for road segments, which is a computationally expensive operation.
Innovation Solution
An electronic device is configured to parallelize the generation of static and dynamic attributes for edges of a graph-structure, with a main processing pipeline handling dynamic attributes and a support processing pipeline handling static attributes, reducing computational time and enhancing decision-making capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the computer system processes data from multiple sensors to generate graph-structures for road segments, then the accuracy of object detection and trajectory prediction is improved, but the computational time and processing complexity increase significantly
Solution Approach 1:
The patent divides the graph-structure generation process into multiple processing pipelines: a main processing pipeline that handles dynamic attributes (sensor data processing, object detection) and a support processing pipeline that handles static attributes (road segment geometry, predefined paths). This segmentation allows parallel execution of independent tasks, reducing overall computational time while maintaining detection accuracy.
Solution Approach 2:
The support processing pipeline pre-generates static attributes for road segments, including graph-structures, potential positions, and transition edges, before the SDC arrives. These pre-computed static attributes are stored and retrieved during real-time operation, eliminating the need to recompute them and significantly reducing real-time processing time.
2Reliability
If the system generates comprehensive graph-structures with multiple potential positions and transitions for the SDC, then the decision-making capability and operational safety are improved, but the device complexity and processing resources increase
Solution Approach 1:
The system separates graph-structure generation into distinct modules: road segment definition, potential position generation, edge transition definition, and attribute computation. Each module handles a specific aspect of the graph-structure, making the overall system more manageable and maintainable while enabling parallel processing to reduce complexity bottlenecks.
Solution Approach 2:
The patent introduces an intermediary support processing pipeline that acts as a mediator between raw sensor data and the main decision-making pipeline. This intermediary pre-processes and structures data into graph-structures with static attributes, reducing the computational burden on the main pipeline and simplifying the overall system architecture.
3Loss of information
If the computer system uses multiple sensor systems (camera, LIDAR, radar) to capture surrounding information, then the completeness and accuracy of environmental perception is improved, but the energy consumption and data processing load increase
Solution Approach 1:
The patent assigns different sensor systems to capture different types of information: camera systems for image data and visual recognition, LIDAR for point cloud data and 3D mapping, and radar for distance and velocity measurement. This segmentation allows each sensor to operate independently and process its own data type, optimizing energy usage while maintaining comprehensive environmental perception.
Solution Approach 2:
The graph-structure framework serves multiple functions simultaneously: it represents road geometry, defines potential vehicle positions, encodes transition possibilities, and integrates data from multiple sensor sources. This multi-functionality reduces the need for separate processing systems for each data type, thereby reducing overall energy consumption and processing load.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method (700) and device (210) for operating a Self-Driving Car (SDC) (220) are disclosed. The device executes in real-time a first processing pipeline including generating static attributes for a plurality of potential positions of the SDC (210) on a road segment (350), and caching the static attributes in association with the respective ones of the plurality of potential positions. The device (210) executes in real-time a second processing pipeline in parallel with the first processing pipeline including generating a graph-structure for operating the SDC (220) on the road segment (350). The generating the graph-structure includes generating dynamic attributes for a given edge of the graph-structure, acquiring from the cache memory static attributes for the given edge of the graph-structure, such that the given edge in the graph-structure is associated with the static attributes generated by the first processing pipeline and with the dynamic attributes generated by the second processing pipeline.