Lidar Scenario Augmentation for Autonomous Driving Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating driving scenarios for autonomous vehicles are limited by the lack of sufficient variety and number, requiring extensive real-world kilometers for training, and simulations are cumbersome and limited to previously encountered scenarios.
Innovation Solution
A method for generating simulation scenarios using LIDAR data that integrates dynamic road users, allowing for the intelligent selection, supplementation, and modification of scenarios, including the use of neural networks to ensure desired statistical distributions and properties, and the generation of simulated sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If real-world driving data is collected to train autonomous driving functions, then the diversity and realism of training scenarios improve, but the time and resources required for data collection increase significantly
Solution Approach 1:
The patent creates virtual copies of real-world driving scenarios by processing recorded sensor data through a simulation environment. The system reconstructs 3D environments from actual sensor recordings and generates synthetic sensor data that mimics real driving conditions, thereby obtaining diverse training scenarios without collecting new real-world data.
Solution Approach 2:
The system performs preliminary processing of sensor data by recording and storing raw sensor data during real drives, then later processes this recorded data through simulation to generate multiple varied scenarios. This allows the diverse scenarios to be created on-demand from pre-recorded data, eliminating the need for time-consuming new data collection.
2Quantity of substance
If simulation environments are used to generate driving scenarios, then the amount of training data can be increased, but the complexity of modeling suitable driving scenarios increases
Solution Approach 1:
The system uses the autonomous vehicle's own recorded sensor data to automatically generate training scenarios. The recorded data serves as both the source material and the ground truth for validation, eliminating the need for manual scenario design and reducing modeling complexity while still producing diverse training data.
Solution Approach 2:
The simulation environment dynamically processes recorded sensor data to generate multiple varied scenarios from the same source material. By applying different transformations, environmental conditions, and parameter variations to the recorded data, the system generates diverse scenarios automatically without requiring complex manual modeling for each scenario.
3Ease of manufacture
If recorded sensor data is replayed in simulation, then existing driving scenarios can be reused, but the variety of scenarios is limited to previously encountered situations
Solution Approach 1:
The system takes recorded sensor data and applies parameter changes to generate varied scenarios. This includes modifying environmental conditions (weather, lighting), object properties (positions, speeds, trajectories), and sensor parameters to create diverse scenarios from the same recorded data, thereby increasing variety while maintaining ease of reuse.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention relates to a computer-implemented method for generating a simulation scenario for a land vehicle, comprising the steps of: - receiving raw data which comprise a plurality of consecutive lidar point clouds; - merging the plurality of lidar point clouds of a particular region in a common coordinate system in order to generate a composite point cloud; - localising and classifying one or more static objects within the composite point cloud; - generating road information on the basis of the composite point cloud, one or more static objects and at least one camera image; - localising and classifying one or more dynamic road users within the plurality of consecutive lidar point clouds and generating trajectories for the road user or users; and - creating a simulation scenario based on the one or more static objects, the road information and the created trajectories for the one or more road users. The generated simulation scenario can be used to test an autonomous driving function.