Semantic Map Sharding for Lower Memory and Network Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles require large, high-definition maps that are typically generated as monolithic data files, which are cumbersome and resource-intensive, necessitating significant network capacity and processing power, especially when loading and processing semantic maps.
Innovation Solution
Semantic map sharding splits high-definition maps into logical chunks called shards, allowing for efficient storage and processing by defining sections within a geographical area and using rules to determine which semantic objects are included in each shard, reducing data size and processing requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If monolithic map data files are used to provide complete geographical coverage, then comprehensive map information is available, but network capacity and processing loads increase significantly
Solution Approach 1:
The patent divides monolithic map data files into smaller logical units called shards. Each shard contains map data for a specific geographical section, allowing the system to load and process only the necessary portions rather than entire monolithic files. This segmentation reduces memory usage and processing loads while maintaining access to comprehensive map information when needed.
2Measurement precision
If high-definition maps with detailed semantic objects are used to enhance autonomous driving safety, then driving safety and navigation accuracy improve, but data size and resource requirements increase
Solution Approach 1:
The patent implements local quality by allowing different shards to have different levels of detail and semantic object richness based on their geographical location and importance. Critical areas with complex intersections or high traffic density can maintain high-definition detailed maps, while less critical areas use simplified representations. This approach maintains navigation accuracy where needed while reducing overall data size and resource requirements.
3Measurement precision
If complete semantic maps are loaded into memory for real-time processing, then real-time navigation and obstacle detection are accurate, but memory consumption increases
Solution Approach 1:
The patent segments map data into shards that can be selectively loaded into memory based on the vehicle's current location and navigation needs. Rather than loading complete semantic maps, the system loads only the relevant shard containing the vehicle's current section, maintaining real-time navigation accuracy while minimizing memory consumption. Shards can be dynamically loaded and unloaded as the vehicle moves through different geographical areas.
Data Source
AI summary
Rules based semantic map sharding is proposed. A method of handling semantic objects for semantic map sharding makes used of rules to define which semantic objects are to be included in a shard. Hereto, semantic objects are defined in a geographical area to be contained in shard data. For the semantic objects, a set of rules is defined comprising a definition of which semantic objects are to be combined into a shard data entry of the shard data and which semantic objects are to be stored in separate shard data entries.