Autonomous Vehicle Driving Style Profile Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles often exhibit driving styles that are perceived as too robotic, lacking the comfort and smoothness associated with human driving, which can be undesirable for passenger experience.
Innovation Solution
An offline machine learning algorithm is employed to analyze human driving data labeled with passenger feedback on comfort, determining an optimal driving style profile that is then integrated into the autonomous vehicle's control and motion planning modules, using techniques like Deep Reinforcement Learning, Random Forest Learning, and Deep Neural Network Learning to simulate and refine driving scenarios such as cut-ins and left-turns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If autonomous vehicle uses automated control system, then automation level is improved, but driving smoothness and passenger comfort deteriorate
Solution Approach 1:
The patent creates multiple human driving style profiles that copy and replicate human driving behaviors. The system captures human driving patterns through sensors and machine learning algorithms, then uses these profiles to guide autonomous vehicle control, making the automated system mimic human smoothness and comfort preferences
Solution Approach 2:
The system dynamically selects and adapts driving profiles based on real-time conditions and passenger feedback. The controller adjusts vehicle control parameters dynamically by choosing from multiple pre-learned profiles or blending them, allowing the system to optimize for comfort while maintaining automation
2Reliability
If autonomous vehicle follows strict safety protocols, then reliability is improved, but passenger comfort deteriorates
Solution Approach 1:
The system changes control parameters by selecting different driving profiles that have been optimized for various conditions. Each profile contains pre-calculated safe yet comfortable parameter ranges for acceleration, braking, and steering, allowing the system to maintain safety while improving comfort
Solution Approach 2:
The system incorporates passenger feedback loops where comfort ratings are collected and used to refine profile selection and blending. This feedback mechanism allows the system to learn passenger preferences and adjust the balance between safety protocols and comfort optimization
Data Source
AI summary
A method for controlling a vehicle includes collecting data about human driving styles; machine learning how the human driver reacts to different traffic scenarios based on the collected data to create a plurality of human driving styles profiles; selecting an optimal driving profile of the plurality of human driving styles profiles, wherein the optimal driving profile is selected based feedback provided by a passenger of the vehicle, the feedback is indicative of a pleasantness of each of the plurality of human driving styles profiles; creating a driving plan based on the optimal driving profile; commanding the vehicle to execute the driving plan in a controlled environment to test the pleasantness of the optimal driving profile; and receiving a pleasantness rating from the passenger of the vehicle while the vehicle executes the driving plan.