GAN-Trained Vehicle Maneuver Software for Robust Control Routines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of automotive software development and maintenance is increasing due to the need for automation in writing code that reacts to various inputs, requiring faster and more robust software solutions, particularly in the context of autonomous vehicles and advanced driving scenarios.
Innovation Solution
A system utilizing a generative adversarial network framework, where a pre-trained neural network processes vehicle maneuver data from sensors to create software routines that are trained and refined through adversarial training, enabling the generation of realistic and robust vehicle control algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional software development methods are used to create vehicle control software, then the software can be developed with established processes, but the development speed is slow and the software lacks robustness in handling complex maneuvers
Solution Approach 1:
The patent replaces traditional mechanical software development processes with an AI-based neural network system. The neural network automatically generates software routines for vehicle maneuvers by learning from training data, substituting the manual coding and testing process with an automated intelligent system that produces more robust software faster.
Solution Approach 2:
The patent changes the fundamental parameters of software development by using adversarial training to optimize the neural network. The generator and discriminator networks iteratively improve the software routines by adjusting parameters such as maneuver trajectories, control inputs, and edge-case scenarios, resulting in software that handles complex situations more reliably.
2Ease of manufacture
If manual software coding is used for vehicle control, then the software can be customized and understood, but the complexity of development and maintenance increases significantly
Solution Approach 1:
The neural network system performs self-service by automatically generating software routines without extensive human intervention. The generator network creates maneuver software autonomously by learning from training data, and the discriminator network automatically validates and refines the generated software, reducing the need for manual coding and simplifying the development process.
Solution Approach 2:
The patent introduces an intermediary AI system between the requirements and the final software product. The neural network acts as a mediator that translates training data into optimized software routines, handling the complexity of software generation and maintenance while providing standardized, high-quality output that is easier to manage than manually coded software.
3Reliability
If extensive training data is used to train the neural network, then the software robustness improves, but the training time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network with extensive training data to establish a robust foundation. The generator and discriminator networks are pre-trained with diverse maneuver data, edge cases, and adversarial examples before deployment, allowing the system to quickly adapt to new scenarios without requiring extensive retraining, thus reducing overall training time while maintaining robustness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure proposes a system (10) for creating a software routine for vehicle maneuvers and a generative adversarial network-based training method thereof. The system (10) comprises an electronic control unit (ECU (102)) in communication with a plurality of vehicle sensors (101), vehicle actuators (104) and at least a pre-trained neural network (103). The pre-trained neural network (103) is trained on dataset comprising multiple vehicle maneuvers in a generative adversarial network framework. The vehicle maneuver data comprises pre-processed values of vehicle operating parameters received from the plurality of vehicle sensors (101).