End-to-End Autonomous Driving Control Without High-Precision Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving technologies face challenges such as error accumulation, coupling issues between prediction and planning, defective structured information representation, and reliance on high-cost, high-precision maps, which limit their effectiveness and applicability.
Innovation Solution
An autonomous driving model with a multimodal encoding layer and decision control layer forms an end-to-end neural network, where perception information is directly responsible for decision-making, reducing error accumulation and eliminating the need for high-precision maps by using lane-level maps and emphasizing perception over mapping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional autonomous driving systems use separate prediction and planning modules with high-precision maps, then comprehensive driving functionality is achieved, but error accumulation occurs and system complexity increases
Solution Approach 1:
The patent merges the prediction module and planning module into a unified end-to-end neural network model. The multimodal encoding layer processes perception information from sensors, and the decision control layer directly generates driving decisions based on this processed information, eliminating the need for separate prediction and planning modules and reducing error accumulation between modules.
Solution Approach 2:
The patent extracts and removes the dependency on high-precision maps from the autonomous driving system. By using lane-level maps combined with real-time perception information from sensors, the system achieves accurate positioning and decision-making without requiring expensive, high-precision map data, thereby simplifying the system while maintaining reliability.
2Measurement precision
If high-precision maps are used for autonomous driving, then accurate position information and road element data are obtained, but cost and system dependency increase
Solution Approach 1:
The patent introduces lane-level maps as an intermediary between the vehicle and the environment, combined with real-time perception information from sensors. This intermediary approach provides accurate position information and road element data without the high cost and rigidity of high-precision maps, while maintaining adaptability to new situations through sensor-based perception.
3Productivity
If multiple processing layers are used between perception and decision-making, then comprehensive information processing is achieved, but error accumulation and processing time increase
Solution Approach 1:
The patent implements an end-to-end neural network model where the multimodal encoding layer continuously processes perception information from sensors and directly feeds the processed features to the decision control layer. This continuous processing flow eliminates discrete processing stages and reduces error accumulation, while maintaining comprehensive information processing through the neural network's inherent feature extraction capabilities.
Data Source
AI summary
An autonomous driving method implemented by using an automatic driving model is provided. The autonomous driving model comprises a multimodal encoding layer and a decision control layer. The autonomous driving method includes: obtaining first input information of the multimodal encoding layer; inputting the first input information into the multimodal encoding layer to obtain an implicit representation corresponding to the first input information output by the multimodal encoding layer; and inputting second input information including the implicit representation into the decision control layer to obtain target autonomous driving strategy information output by the decision control layer.