Autonomous Driving Path Training for Unseen Scenario Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion planning systems for autonomous vehicles struggle to adapt to driving scenarios outside the scope of expert data, leading to errors and instability in path generation and control systems.
Innovation Solution
A data augmentation method is employed to enhance training data by adding noise to vehicle parameters, using a vehicle dynamics model and objective function to generate optimal paths, combined with open-loop simulation and closed-loop reinforcement training using the BC-SAC algorithm to improve the driving path planner and controller.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a motion planning model is trained based on expert data set, then the model can infer driving trajectories accurately for scenarios within the expert data, but the model fails to correctly modify or infer driving trajectories for situations outside the expert data scope
Solution Approach 1:
The patent applies preliminary action by pre-training the motion planning model on expert data to establish baseline trajectory inference capabilities, then subsequently training on augmented data that includes noisy trajectories to prepare the model for unseen scenarios before actual deployment
Solution Approach 2:
The patent changes the parameters of training data by introducing noise to the expert trajectories, transforming clean expert data into augmented training data with varied trajectories. This parameter change enables the model to learn robust trajectory modification capabilities for scenarios outside the original expert data scope
2Adaptability or versatility
If data augmentation with noise is applied to training data, then the model's generalization to unseen scenarios improves, but the training data quality and consistency may deteriorate
Solution Approach 1:
The patent converts the potential harm of noisy, inconsistent data into a benefit by deliberately adding controlled noise to expert trajectories. This noise, which initially appears to degrade data quality, actually serves as valuable training examples that teach the model how to adapt to variations and unseen scenarios, improving generalization capability
3Reliability
If open-loop simulation training and closed-loop reinforcement training are used to train the driving path planner and controller, then the stability and performance of autonomous driving systems improve, but the training complexity and computational resources increase
Solution Approach 1:
The patent segments the training process into two distinct phases: open-loop simulation training for the driving path planner and closed-loop reinforcement training for the driving controller. This segmentation allows each component to be trained with appropriate methods and metrics, improving overall system stability while making the complex training process more manageable and systematic
Data Source
AI summary
A driving path optimization training method of a vehicle includes: receiving a first data set including a driving path and an associated driving environment information; generating a second data set from the first data by performing data augmentation on the first data; training a driving path planner based on the second data set; and training a driving controller based on a training result of the training of the driving path planner.