Autonomous Vehicle Driving Control With Reinforcement Learning Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle control systems face responsiveness and precision issues due to discrepancies between actual and reference trajectories, requiring complex and synchronized deliberative and intuitive decision-making systems, and frequent relearning processes that affect performance.
Innovation Solution
A multi-evolutionary decision-making method using fuzzy logic and reinforcement learning to combine real-time data from vehicle, road, and environmental parameters, adapting vehicle control based on past experiences to mimic human driving behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a deliberative or intuitive method is used to generate control commands, then decision-making adaptability is improved, but system complexity increases due to requiring two parallel synchronized systems
Solution Approach 1:
The patent segments the decision-making process into two distinct modes (deliberative and intuitive) that are selectively activated based on situation complexity. This segmentation allows each mode to be optimized independently while reducing the need for continuous synchronization between parallel systems, thereby maintaining adaptability while managing system complexity.
2Measurement precision
If complete relearning is performed with each new data processing, then learning accuracy is improved, but processing time increases
Solution Approach 1:
The patent implements preliminary action by pre-processing and organizing data during idle periods or between decision cycles, preparing learning datasets in advance. This allows the system to perform faster inference during critical decision moments while maintaining high learning accuracy through pre-computed models and pre-organized data structures.
Solution Approach 2:
The system employs periodic action by performing complete relearning at scheduled intervals or when confidence thresholds are met, rather than continuously. Between these periodic relearning events, the system uses lighter-weight inference mechanisms, thus achieving high accuracy when needed while minimizing time loss through rhythmic, efficient learning cycles.
3Productivity
If the vehicle follows the reference trajectory calculated in real time, then route optimization is improved, but trajectory tracking precision decreases due to constraints not accounted for in calculation
Solution Approach 1:
The patent implements feedback by continuously monitoring the discrepancy between the reference trajectory and actual vehicle position, then adjusting control commands in real-time to compensate for unmodeled constraints. This closed-loop feedback mechanism maintains both route optimization and trajectory tracking precision by adapting to actual vehicle behavior and environmental conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for assisting with driving a motor vehicle, comprising a step (E3) of generating a control response (Rep1) to be made with respect to a current situation, and a step (E4) of sending a command (K1) to an actuator in order to control the vehicle (10). In accordance with an item of quality information regarding the result of the command, the method further comprises a step (E5) of generating, by reinforcement learning, a new control response (Rep2), said new control response (Rep2) being optimised with respect to the current situation.