Self-driving Simulation Rendering via Tree Model Cropping

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing self-driving simulation software inefficiently renders scenes, wasting resources by rendering unnecessary information and struggling with multi-sensor simulations due to low rendering efficiency and underutilization of GPU resources.

Innovation Solution

The method employs a tree model for efficient cropping of the effective area, dividing scene simulation rendering into two stages, and using a multi-thread parallel mode to allocate resources effectively, ensuring only valid information is rendered and maximizing resource utilization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If all objects within a certain distance around the present vehicle are acquired for rendering, then complete scene information is obtained, but resource waste occurs due to rendering unnecessary information

Engineering Contradiction:
Improvescene information completenessVSAvoidrendering resource waste
Core Design Contradiction:
Loss of informationVSLoss of energy

Solution Approach 1:

The patent extracts and processes only the necessary effective area from the complete scene using a tree model. By representing the scene as a hierarchical tree structure and identifying the effective area corresponding to sensor positions and directions, the system extracts only the relevant portion of the scene for rendering, eliminating unnecessary rendering operations while preserving all required information.

Inventive Principle:
Principle #2Taking out (Extraction)

2Adaptability or versatility

If multiple viewports are rendered simultaneously for multi-sensor simulation, then all sensor data is captured, but rendering efficiency decreases and GPU resources are underutilized

Engineering Contradiction:
Improvemulti-sensor simulation capabilityVSAvoidrendering efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent segments the rendering process into two distinct stages: a first stage that processes common effective area information shared by multiple sensors, and a second stage that handles sensor-specific rendering. This segmentation allows the system to render the common portion once and reuse it across multiple viewports, significantly improving rendering efficiency and GPU resource utilization while maintaining full multi-sensor simulation capability.

Inventive Principle:
Principle #1Segmentation

3Device complexity

If scene rendering is performed without dividing into stages, then the process is simple, but rendering speed is slow and real-time simulation cannot be achieved

Engineering Contradiction:
Improverendering process complexityVSAvoidrendering speed
Core Design Contradiction:
Device complexityVSSpeed

Solution Approach 1:

The patent implements preliminary action by performing the effective area cropping operation in the first stage before the actual sensor-specific rendering in the second stage. By pre-processing and identifying the effective area using the tree model in advance, the system prepares optimized data structures and reduces the computational burden of the subsequent rendering stage, thereby achieving real-time rendering speed while maintaining manageable process complexity through clear stage separation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS12315078B2Self-driving simulation rendering method and apparatus, device, and readable medium
Publication Date: 2025.05.27 INSPUR SUZHOU INTELLIGENT TECH CO LTD
  • US12315078B2 patent drawing
  • US12315078B2 patent drawing

AI summary

A self-driving simulation rendering method and apparatus, a computer device, and a readable storage medium wherein the method includes: loading a scene file, performing scene modeling based on a tree model, and establishing a static spatial index (S1); setting a dynamic traffic flow to generate a dynamic scene, and setting a position and direction of a sensor (S2); and creating a first thread and a second thread, scheduling the first thread to crop each frame of the dynamic scene based on the tree model, and scheduling the second thread to render, based on the position and direction of the sensor, each frame cropped by the first thread and output the rendered frame (S3). Modeling and efficient cropping of an effective area are carried out by means of a tree model, which ensures that only valid information is rendered, thereby reducing resource waste.