Image-Based Predictive Vehicle Control for Sharp Corner Steering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for steering control in autonomous vehicles using digital images face challenges, particularly in navigating sharp corners safely, as they often rely on lane detection which can be unreliable due to road diversity, weather conditions, and lack of pre-built maps, and require extensive sensor data processing or costly hardware.
Innovation Solution
The method involves generating predictions for future lane centeredness and road angle using General Value Functions (GVFs) implemented by neural networks, allowing for smoother steering and speed control by anticipating road curvature and adjusting vehicle actions accordingly, without relying on expensive sensors or detailed maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If more sensors and high-quality sensors are included in the autonomous vehicle, then the ADS can better control operation of the vehicle, but it becomes too costly and processor-intensive
Solution Approach 1:
The patent extracts and focuses on a single critical sensor type (camera) that captures digital images, eliminating the need for multiple sensor types. The system processes only image data to generate steering control, removing unnecessary sensors and processing loads while maintaining effective autonomous control
Solution Approach 2:
The patent uses standard digital cameras and image processing techniques that are inexpensive and widely available, replacing costly specialized sensors. The system achieves reliable control using affordable image capture devices and computational methods
2Ease of operation
If conventional lane detection methods are used for steering control, then the system can generate steering commands, but it fails in diverse road conditions, weather, and without pre-built maps
Solution Approach 1:
The patent trains neural networks in advance using large datasets of digital images with annotated lane information. This preliminary training enables the system to generalize to diverse road conditions, weather scenarios, and unseen environments without requiring pre-built maps or real-time lane detection
Solution Approach 2:
The patent replaces traditional computer vision algorithms that mechanically detect lane markings with learned neural network models. These models learn abstract representations of drivable paths from training data, enabling robust steering control that works across diverse conditions without relying on visible lane markings
3Device complexity
If digital images alone are used for steering control, then the system reduces sensor cost and processing requirements, but it becomes difficult to steer around sharp corners safely
Solution Approach 1:
The neural networks are pre-trained on extensive datasets that include sharp corners and challenging navigation scenarios. This preliminary learning enables the system to anticipate and safely navigate sharp corners using only image data, without requiring additional sensors or complex processing
Solution Approach 2:
The patent transforms the control problem from direct image-to-steering mapping to a learned representation space where the neural networks extract meaningful features. This dimensional transformation enables the system to understand road geometry and navigate sharp corners safely using only visual information
Data Source
AI summary
Methods and systems for predictive control of an autonomous vehicle are described. Predictions of lane centeredness and road angle are generated based on data collected by sensors on the autonomous vehicle and are combined to determine a state of the vehicle that are then used to generate vehicle actions for steering control and speed control of the autonomous vehicle.