Vehicle Path Planning Using Historical Obstacle Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining a drivable path for vehicles are prone to false positive detections by sensors, leading to inconvenient and potentially dangerous situations, especially when recurring obstacles like road cracks are misclassified as non-overdrivable, and rely heavily on reliable sensor data and processing power, which is not efficient.
Innovation Solution
A computer-implemented method that uses sensor data to determine confidence scores for obstacles, storing and updating these scores based on previous encounters, allowing the vehicle to learn which obstacles are overdrivable and include them in the path planning, while excluding non-overdrivable ones, using a combination of sensor data, AI engines, and data storage for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data is used to detect obstacles in real-time, then the vehicle can navigate safely, but false positive detections occur leading to unnecessary path adjustments
Solution Approach 1:
The system performs preliminary actions by storing historical sensor data and obstacle information in advance. When an obstacle is detected, the system checks historical data from previous encounters at the same location to verify if the obstacle is real or a false positive, thereby reducing unnecessary path adjustments
Solution Approach 2:
The system implements feedback by comparing current sensor detections with historical data from previous rides. If an obstacle detected in real-time matches historical false positive patterns, the system corrects the detection and prevents unnecessary path adjustments, continuously improving detection accuracy
2Object-affected harmful factors
If the drivable path is adjusted to avoid detected obstacles, then collision avoidance is achieved, but dangerous situations may arise from sudden maneuvers
Solution Approach 1:
The system performs preliminary verification by checking historical data before adjusting the drivable path. If historical information indicates the detected obstacle is a false positive (e.g., road cracks or bumps that are actually drivable), the system maintains the original path instead of making sudden steering maneuvers
Solution Approach 2:
The system uses feedback from historical ride data to validate real-time obstacle detections. When the feedback indicates the obstacle is not a real threat, the system suppresses unnecessary path adjustments, ensuring smooth and safe vehicle operation
3Reliability
If neural network architectures are used to estimate predictive uncertainty, then false positive reduction is achieved, but processing power requirements increase
Solution Approach 1:
The system performs preliminary processing by storing sensor data and obstacle information in historical databases during previous rides. This pre-processing allows real-time verification using stored data rather than requiring intensive neural network computations during actual driving, reducing processing power requirements
Solution Approach 2:
The system uses simplified verification methods that replicate the essence of neural network uncertainty estimation without requiring full neural network execution. By comparing current detections with historical patterns, the system achieves false positive reduction with lower computational overhead
4Measurement precision
If the vehicle learns from previous encounters to classify obstacles, then detection accuracy improves, but data storage requirements increase
Solution Approach 1:
The system extracts only the essential and relevant features from sensor data for storage, such as obstacle characteristics, location, and detection confidence levels, rather than storing complete raw sensor datasets. This extraction approach maintains classification accuracy while significantly reducing data storage requirements
Solution Approach 2:
Instead of storing all possible obstacle data and filtering later, the system inverts the approach by storing only verified false positive patterns and using these compact references for quick comparison during real-time operation, reducing storage needs while maintaining accuracy
Data Source
AI summary
A computer-implemented method for determining a drivable path for a vehicle during a ride includes obtaining sensor data. The sensor data indicates a perceived obstacle associated with a geographical location. The method includes retrieving, from a data storage, data associated with the perceived obstacle. The data includes a confidence score. The method includes, in response to the confidence score being above a threshold, determining the perceived obstacle as overdrivable for the vehicle.