Spline-Based Exit Path Prediction at Unmarked Intersections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques for vehicle control in autonomous vehicles are less effective in complicated traffic environments, particularly at complex intersections where interactions between vehicles are complex, and fail to accurately predict the likelihood of a vehicle choosing a given exit lane.
Innovation Solution
A system that uses a vehicle behavior planning system with a neural network to detect vehicles entering an unmarked road segment, determine available exit paths, predict the exit path taken by the detected vehicle, and plan a trajectory for the ego vehicle to navigate the intersection based on calculated splines and sensor data from cameras, LIDAR, and other sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning techniques are used for vehicle control, then the system is simple to implement, but the prediction accuracy of exit lane choices deteriorates in complicated traffic environments
Solution Approach 1:
The system segments the complex task of exit lane prediction into multiple specialized modules: a spline generation module that creates candidate paths, a curvature analysis module that evaluates path characteristics, and a prediction module that selects the most likely exit lane. This segmentation allows each module to focus on a specific aspect of the problem, improving overall prediction accuracy while managing system complexity through modular design.
Solution Approach 2:
The system introduces a new dimensional approach by generating spline curves as intermediate representations of possible vehicle paths. Instead of directly predicting exit lane choices from raw sensor data, the system transforms the problem into analyzing path curvatures and geometries in the spline domain, which provides richer geometric insights for predicting vehicle intentions at unmarked road segments.
2Measurement precision
If the system analyzes multiple exit paths using splines, then the prediction accuracy improves, but the computational time increases
Solution Approach 1:
The system generates multiple spline paths (excessive action) to ensure comprehensive coverage of possible vehicle trajectories, but then applies curvature-based filtering to eliminate unlikely paths early in the analysis. This allows the system to maintain high prediction accuracy by considering multiple possibilities while reducing computational burden by discarding implausible paths before detailed analysis.
Solution Approach 2:
The system performs preliminary spline generation and curvature calculation for all candidate paths before the actual prediction decision is made. By pre-computing path geometries and curvature characteristics, the system prepares the data structure in advance, enabling faster final prediction when needed and reducing real-time computational requirements.
3Adaptability or versatility
If the system uses curvature-based spline ranking, then the ability to handle unmarked road segments improves, but the complexity of path determination increases
Solution Approach 1:
The system changes the parameter space for path evaluation by using curvature as the primary ranking criterion instead of relying on traditional lane marking detection. This parameter transformation allows the system to adapt to unmarked road segments where conventional lane-based methods fail, as curvature provides a continuous geometric descriptor that works regardless of the presence of road markings.
Solution Approach 2:
The system replaces the mechanical/visual system of detecting lane markings with a geometric field-based approach using spline curves and curvature analysis. Instead of relying on discrete visual cues from road markings, the system uses continuous mathematical representations of vehicle paths, substituting visual pattern recognition with geometric field analysis that is more adaptable to unmarked segments.
Data Source
AI summary
A method for vehicle behavior prediction is described. The method includes detecting a vehicle entering an unmarked road segment. The method also includes determining a set of exit paths available for the detected vehicle to reach different exit lanes of the unmarked road segment. The method further includes predicting an exit path taken by the detected vehicle from the set of exit paths available for the detected vehicle to reach the different exit lanes of the unmarked road segment. The method also includes planning a trajectory of an ego vehicle according to the predicted exit path taken by the detected vehicle to reach an exit lane of the unmarked road segment.