Graph-Based Synthetic Training Data Generation for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning algorithms in autonomous driving require vast amounts of training data to accurately predict driving tasks, making it resource-intensive and challenging to capture all possible scenarios, necessitating an improved method for generating additional training data.
Innovation Solution
The method involves transforming known training data into a graph structure, where nodes represent objects, and modifications such as dropping, perturbing, mirroring, masking, duplicating, or moving nodes are applied to generate additional training data, allowing for the creation of diverse scenarios without requiring extensive resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If vast amounts of training data are collected to accurately predict all possible driving scenarios, then prediction accuracy is improved, but storage space and computational resources are excessively consumed
Solution Approach 1:
The patent creates copies of existing training data instances by applying graph transformations (node dropping, perturbation, mirroring, masking, duplicating, moving) to generate synthetic training examples. This copying approach allows the system to expand training data coverage without collecting or storing additional real-world data, thereby improving prediction accuracy while avoiding proportional increases in storage and computational resources
Solution Approach 2:
The patent transforms training data into graph structures and applies parameter changes through various modification operations (dropping nodes, perturbing positions, mirroring orientations, masking features, duplicating objects, moving locations). These parameter changes generate diverse training scenarios from limited original data, enabling the model to learn from varied situations without requiring vast amounts of stored training data
2Adaptability or versatility
If training data is transformed into graph structure and modified through multiple operations, then diversity of training scenarios is improved, but complexity of data processing is increased
Solution Approach 1:
The patent segments training data into discrete graph components (nodes representing objects, edges representing relationships). This segmentation allows independent modification of individual elements through operations like node dropping, perturbation, mirroring, masking, duplicating, and moving. By working with segmented graph elements rather than complete scenes, the system achieves high scenario diversity while keeping processing complexity manageable through modular operations
Solution Approach 2:
The patent transforms traditional structured data into graph structure representation, adding a new dimensional framework for data organization and manipulation. This graph-based dimension enables versatile transformations (topological modifications, spatial perturbations, relational changes) that generate diverse scenarios. The graph structure serves as an intermediate representation that simplifies complex modifications compared to direct manipulation of raw sensor data
Data Source
AI summary
A method for generating additional training data for training a machine learning algorithm is disclosed. The method includes (i) providing training data for training the machine learning algorithm, wherein the training data includes labeled sensor data from at least one sensor, (ii) transforming the training data for training the machine learning algorithm in a graph structure, wherein nodes in the graph structure represent objects represented in the corresponding sensor data, and wherein a starting node of the graph structure represents the position of the at least one sensor with respect to the objects represented in the corresponding sensor data, and (iii) generating additional training data for training the machine learning model by modifying the graph structure.