Obstacle Trajectory Simulation Using Gaussian Perception Deviations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating obstacles in unmanned simulation scenes are manual and fail to accurately reflect the dynamic and unstable perception results of actual driving environments, leading to unrealistic testing conditions for autonomous vehicles.
Innovation Solution
A method that uses Gaussian distribution information based on actual perception performance to adjust initial motion trajectory sequences of simulated obstacles, incorporating position and contour deviations, and residence periods, to create a more realistic three-dimensional scene map for simulation testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual methods are used to construct obstacle information in the simulation scene, then the construction process is simple and straightforward, but the accuracy and realism of the simulated obstacle positions do not reflect actual perception performance
Solution Approach 1:
The patent applies parameter changes by introducing Gaussian distribution parameters (expectation value and variance value) to model obstacle position deviations. Instead of using fixed manual positions, the system dynamically adjusts obstacle positions based on statistical parameters that reflect actual perception algorithm performance, thereby improving position accuracy while maintaining manageable complexity through parameterized control.
Solution Approach 2:
The patent implements feedback by using actual perception performance data to determine Gaussian distribution parameters, which then guide the adjustment of obstacle positions in the simulation scene. This closed-loop approach ensures that the simulated obstacle positions continuously reflect real-world perception characteristics, improving realism without requiring complete manual reconstruction of the scene.
2Reliability
If actual vehicle testing is performed to verify unmanned vehicle function, then good verification results can be achieved, but the cost and risk are extremely high
Solution Approach 1:
The patent applies the copying principle by creating a virtual simulation scene that replicates real-world driving scenarios and perception characteristics. Instead of testing directly on actual vehicles in real environments, the system copies essential features (obstacle positions, motion trajectories, perception performance) into a simulated environment, maintaining verification reliability while eliminating the high costs and risks of physical testing.
Solution Approach 2:
The patent implements preliminary action by constructing and validating the simulation scene before actual vehicle testing. The Gaussian distribution parameters are determined based on pre-collected perception performance data, and obstacle information is prepared in advance. This allows the simulation to be ready for testing ahead of time, ensuring reliability while avoiding the need for costly and risky real-world trial-and-error testing.
3Adaptability or versatility
If manual construction of obstacles is used in the simulation scene, then the process is easy to implement, but it fails to reflect the instability and variability of real-world perception results
Solution Approach 1:
The patent applies dynamics by transforming static manual obstacle positions into dynamic positions that vary according to Gaussian distribution. Instead of fixed obstacle locations, the system generates positions that fluctuate within statistically defined ranges, reflecting the instability and variability of real perception results. This dynamic approach improves adaptability while the automated generation process maintains operational ease.
Data Source
AI summary
A method for simulating an obstacle in an unmanned simulation scene includes: for obstacle information in a three-dimensional scene map, determining Gaussian distribution information of obstacle position detected by using a perception algorithm to be tested based on actual perception performance of the perception algorithm; adjusting each position of the simulated obstacle in the initial motion trajectory sequence, such that a position deviation between each adjusted position point in the target motion trajectory sequence and the corresponding position point in the initial motion trajectory sequence follows a Gaussian distribution; and adding the target motion trajectory sequence of the simulated obstacle to the three-dimensional scene map.