Trajectory Prediction Using Polynomial Coefficients and Variance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object tracking systems in autonomously driving vehicles lack reliability and efficiency, particularly in processing radar data for predicting object trajectories and handling uncertainty in motion forecasting.
Innovation Solution
A computer-implemented method using a recurrent convolutional one-stage feature pyramid network that combines radar data processing with polynomial functions for trajectory prediction, incorporating variance estimation and multivariate normal distributions to improve tracking accuracy and handle uncertainty, employing a two-phase training approach to optimize network performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If polynomial functions are used for trajectory prediction, then trajectory prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent transforms the trajectory prediction problem by changing parameters from raw radar measurements to polynomial coefficients. The neural network outputs coefficients of polynomial functions (e.g., constant term, linear term, quadratic term) that parameterize the trajectory, allowing continuous and smooth predictions while maintaining computational efficiency through standard linear algebra operations.
Solution Approach 2:
The patent moves from predicting trajectories directly in spatial coordinates to predicting trajectories in the polynomial coefficient space. This dimensional transformation allows the use of simple linear regression and standard neural network layers to achieve complex trajectory modeling, reducing computational complexity while improving accuracy.
2Reliability
If variance estimation is incorporated, then uncertainty handling is improved, but model complexity increases
Solution Approach 1:
The patent segments the output of the neural network into separate components: mean trajectory parameters and variance parameters. The network independently predicts both the polynomial coefficients for the mean trajectory and the variance parameters, allowing explicit uncertainty quantification without requiring complex probabilistic models. This segmentation simplifies the overall model architecture while improving reliability.
Solution Approach 2:
The patent introduces variance parameters as an intermediary between the neural network predictions and the final trajectory estimation. These variance parameters act as a mediator that quantifies uncertainty in the trajectory prediction, allowing the system to handle uncertain radar measurements and improve overall reliability without complicating the core prediction mechanism.
3Adaptability or versatility
If multi-object tracking is performed, then tracking comprehensive is improved, but processing time increases
Solution Approach 1:
The patent merges multiple object tracking tasks into a single unified neural network framework. The network simultaneously processes radar data for multiple objects, predicts trajectories for all objects, and estimates variances for all objects in one forward pass. This merging approach improves comprehensive tracking capability while avoiding the cumulative processing time that would result from sequential processing of individual objects.
Data Source
AI summary
The disclosure includes a computer-implemented method for predicting trajectory data of an object including: acquiring radar data of the object; determining a parametrization of the trajectory data of the object based on the radar data; and determining a variance of the trajectory data of the object based on the radar data. The trajectory data of the object includes a position of the object and a direction of the object. The parametrization includes a plurality of parameters and a polynomial of a pre-determined degree. The parameters include a plurality of coefficients related to elements of a basis of the polynomial space of polynomials of the pre-determined degree.