Autonomous Vehicle Risk Model Training from Driver Interventions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving modes in vehicles face issues of driving insecurity due to inadequate driver intervention in unexpected situations, leading to potential accidents and increased maintenance costs.
Innovation Solution
A vehicle risk model training system that collects and analyzes actual and planned driving data, identifies manual intervention data, determines risk levels, and trains the vehicle risk model to improve driving safety by incorporating driver interventions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If autonomous driving mode is implemented with basic sensor-based environment perception, then driving operations are reduced and driving experience is improved, but driving insecurity occurs due to inadequate driver intervention capability in unexpected situations
Solution Approach 1:
The system collects actual driving data from sensors and compares it with planned driving data to identify deviations. When the driver manually intervenes to correct unexpected situations, this intervention data is captured and fed back to retrain the risk model, creating a continuous improvement loop that enhances both automation reliability and safety
Solution Approach 2:
The risk model is proactively trained using historical manual intervention data before actual autonomous driving operations. By pre-learning from past driver corrections, the system prepares the AI to anticipate and respond to similar situations, reducing the need for manual intervention while maintaining safety
2Reliability
If the vehicle risk model is trained extensively with manual intervention data to improve safety, then driving safety is enhanced, but the system complexity and data processing requirements increase
Solution Approach 1:
The system extracts only the critical manual intervention data where actual driving data deviated from planned driving data. By filtering and selecting only the relevant correction cases rather than processing all driving data, the training process becomes more manageable while maintaining safety improvements
Solution Approach 2:
The training process is divided into distinct phases: collecting actual driving data, comparing with planned data, identifying manual intervention cases, and retraining the risk model. This segmentation allows each step to be optimized independently and simplifies the overall complex process
Data Source
AI summary
A vehicle risk model training method applied to an autonomous vehicle. The vehicle risk model training method comprises obtaining driving data from the autonomous vehicle and environmental perception data, wherein the driving data comprises actual driving data and planned driving data generated by an vehicle risk model of the autonomous vehicle, setting data differences between the actual driving data and the planned driving data as a manual intervention driving data, determining a first perception data matching with the manual intervention driving data in the environmental perception data, and determining risk level information matching with the manual intervention driving data according to the first perception data, and training the vehicle risk model based on the risk level information, the first perception data, the planned driving data, and the manual intervention driving data. A vehicle device and a path planning method are also disclosed.