Sensor Fusion Using Deep Neural Networks for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for sensor fusion in autonomous vehicles require manual design of heuristics and parameters, which is time-consuming and resource-intensive, and needs frequent updates with changes in sensor configurations, making them inefficient and inaccurate.
Innovation Solution
The use of machine learning models, specifically deep neural networks, to process and fuse sensor data from various types of sensors without manual design of heuristics and parameters, allowing for automatic updates and improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual heuristics and parameters are used for sensor data fusion, then the system can be implemented with traditional processing methods, but the development time and computing resources required are excessive
Solution Approach 1:
The patent replaces manual heuristic-based processing systems with a machine learning model that automatically learns optimal fusion strategies from training data. The ML model substitutes the mechanical process of manual parameter tuning and heuristic design, enabling the system to achieve accurate sensor data fusion without extensive developer intervention or manual configuration.
2Reliability
If manual heuristics and parameters are used for sensor data fusion, then the system can process sensor data using traditional methods, but the computing resources required are excessive
Solution Approach 1:
The patent substitutes computationally intensive manual heuristic processing with a trained machine learning model that performs fusion operations more efficiently. The ML model, once trained, executes predictions with lower computational overhead compared to traditional multi-step heuristic-based systems, reducing the computing resources required for real-time sensor data fusion.
3Ease of manufacture
If manual heuristics and parameters are used for sensor data fusion, then the system can be implemented with traditional processing pipelines, but frequent updates are needed when sensors are added or removed
Solution Approach 1:
The patent implements a dynamic system where the machine learning model can adapt to different sensor configurations without requiring manual reconfiguration of heuristics and parameters. The model learns from training data that encompasses various sensor setups, enabling it to dynamically adjust its fusion strategy based on the actual sensors present in the system, thus providing both ease of implementation and high adaptability.
Solution Approach 2:
The machine learning model serves as a universal fusion mechanism that can handle multiple sensor types and configurations through a single unified approach. Rather than requiring separate heuristic rules for each sensor combination, the ML model generalizes across different sensor configurations, making the system both easy to implement and highly versatile.
4Reliability
If traditional multi-step processing is used for sensor fusion, then the system can follow established methods, but the number of heuristics and parameters required increases complexity
Solution Approach 1:
The patent merges multiple separate heuristic processing steps into a single integrated machine learning model. Instead of implementing numerous independent heuristics and parameters for different fusion operations, the ML model combines these functions into one unified system that performs association, tracking, and fusion operations simultaneously, thereby maintaining high tracking accuracy while significantly reducing system complexity.
Data Source
AI summary
In various examples, sensor fusion for autonomous or semi-autonomous systems and applications is described. Systems and methods are disclosed that use data pipelines to process sensor data generated using different types of sensors (e.g., image sensors, RADAR sensors, LiDAR sensors, etc.) in order to generate first data representing information associated with objects surrounding a vehicle. For instance, the information may represent values for parameters associated with the objects, such as locations of the objects, dimensions of the objects, velocities of the objects, orientations of the objects, classifications of the objects, and/or any of parameters. The systems and methods may then process the first data using one or more machine learning models (e.g., one or more deep neural networks) that are trained to fuse the information (e.g., the parameters) and output second data representing final information associated with the objects. The fused output may then be used to perform downstream operations.