Neural Network Vehicle Behavior Estimation for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current information processing systems for vehicles, such as those described in PTL 1, fail to accurately estimate vehicle driving operations, leading to issues like driver unease during autonomous driving, inability to react to changing situations, and difficulties in switching between autonomous and manual driving modes.
Innovation Solution
An information processing system that includes a detector for vehicle environment and state, a behavior learning unit using neural networks to learn relationships between environment and behavior, and an estimation unit to predict vehicle behavior, with features like transfer learning for specific drivers and scene-specific neural networks to enhance prediction accuracy and prevent inappropriate behavior notifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a driving control device is used to allow visual recognition of operational state during autonomous driving, then driver awareness of vehicle behavior is improved, but the ability to accurately estimate driving conduct is insufficient
Solution Approach 1:
The patent replaces traditional rule-based or mechanical estimation methods with a neural network-based system. The neural network learns complex patterns from historical driving data to accurately estimate driving conduct, overcoming the limitations of conventional approaches while maintaining driver awareness through visual feedback.
Solution Approach 2:
The system uses the vehicle's own historical driving data and sensor information to train and improve its estimation capabilities. By continuously learning from its own operational data, the system enhances its ability to estimate driving conduct without requiring external intervention or complex additional hardware.
2Extent of automation
If autonomous driving control is implemented, then driving automation is improved, but driver trust and comfort during mode transitions deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the estimated driving conduct is visually presented to the driver before execution. This allows the driver to understand and anticipate the autonomous system's intentions, building trust and reducing anxiety during autonomous driving and mode transitions. The feedback loop creates transparency in the automation's decision-making process.
Solution Approach 2:
The system performs preliminary estimation of driving conduct and presents it to the driver before the actual autonomous action is executed. This advance notification allows the driver to mentally prepare and understand the upcoming maneuver, improving comfort and trust during autonomous driving operations.
3Device complexity
If traditional behavior estimation methods are used, then system complexity is reduced, but the ability to handle diverse driving scenarios and drivers deteriorates
Solution Approach 1:
The patent changes the fundamental parameter of behavior estimation from rule-based logic to neural network-based probabilistic modeling. This allows the system to adapt to diverse driving scenarios and individual driver characteristics by learning from data, rather than relying on pre-programmed rules that cannot accommodate variability.
Solution Approach 2:
The neural network-based estimation system serves multiple functions: it estimates driving conduct for different drivers, adapts to various driving scenarios, and provides the basis for visual feedback to the driver. This single versatile approach replaces multiple specialized systems, achieving adaptability without proportionally increasing complexity.
Data Source
AI summary
An information processing system that appropriately estimates a driving conduct includes: a detector that detects a vehicle environment state, which is at least one of surroundings of a vehicle and a driving state of the vehicle; a behavior learning unit configured to cause a neural network to learn a relationship between the vehicle environment state detected by the detector and a behavior of the vehicle implemented after the vehicle environment state; and a behavior estimation unit configured to estimate a behavior of the vehicle by inputting, into the neural network that learned, the vehicle environment state detected at a current point in time by the detector.