Transforming Automotive Policy Creation with Generative AI

Revolutionizing Data Utilization in Software-Defined Vehicles

Overview of Sonatus’s AI-Powered Solutions

Addressing Challenges in Data Collection and Automation

Key Metrics for Evaluating Success

Innovative Solution Architecture for Policy Generation

Highlights of the Multimodal Approach to Task Execution

Conclusion: Unlocking Efficiency in Policy Creation with Generative AI

Transforming Vehicle Data Management: The Sonatus and AWS Partnership

In today’s automotive landscape, data is more than just numbers; it is the heart of innovation for Original Equipment Manufacturers (OEMs). As the drive towards Software-Defined Vehicles (SDVs) accelerates, the role of sophisticated data management systems becomes increasingly vital. Sonatus is at the forefront of this evolution with its Collector AI and Automator AI products, designed to enhance vehicle data utilization and automate vehicle functions seamlessly.

The Challenge of Data and Automation in Modern Vehicles

As vehicles become more digitalized, the complexity of managing vehicle data grows. OEM engineers face a daunting task of selecting from thousands of data signals to support various use cases. Additionally, automating vehicle functions requires a deep understanding of events and signals—skills that not all OEM users possess.

This is where Sonatus’s Collector AI and Automator AI shine. Collector AI allows for data gathering without any modifications to vehicle electronics, simplifying the data collection policies. Meanwhile, Automator AI offers a no-code approach to automation, which, though intuitive, can still be challenging for users unfamiliar with the underlying systems.

Innovating with Natural Language Processing

Recognizing these challenges, Sonatus has partnered with the AWS Generative AI Innovation Center, harnessing generative AI to create a natural language interface capable of streamlining the policy generation process. This innovation promises to cut down policy creation time from days to mere minutes, making it accessible to engineers and non-experts alike.

How the System Works

The collaboration has produced a system that automates policy generation by understanding user queries expressed in natural language. The system breaks down these queries into manageable components, extracts necessary data, translates it into structured representations, and compiles it into actionable vehicle policies.

Key Components of the Solution:

1. Entity Extraction: User queries are parsed to identify triggers and actions. For instance, a request to “lock the doors when the car is moving” is transformed into a structured output indicating specific vehicle signals.

2. Signal Translation: The correct signals are identified and translated into a predefined format, adhering to the Vehicle Signal Specification (VSS) standards.

3. Automated Policy Generation: The assembled data is validated against established schemas, ensuring quality and consistency.

Overcoming Challenges

Throughout the implementation process, the team faced various hurdles, such as:

• Complex Event Structures: Different vehicle models and policies have varied representations, necessitating a flexible generation approach.
• Labeled Data Limitations: The scarcity of labeled data for mapping natural language to specific policies posed a challenge.
• Quality Assurance: Ensuring the accuracy and consistency of generated policies is paramount for building trust in the system.

Metrics for Success

To gauge the efficacy of their solution, Sonatus established several metrics, both business and technical, including reduced policy generation time, expanded user bases, and accuracy of generated policies.

A Multi-Agent Approach for Precision

One of the standout features of this policy generation system is its use of a multi-agent approach. By incorporating two agents—ReasoningAgent and JudgeAgent—the system iteratively proposes and refines signal names based on user input and knowledge bases, ensuring that the correct context is always maintained.

Furthermore, by merging certain tasks and calls, the system optimizes performance and reduces latency, enabling faster and more efficient operations.

Conclusion

The partnership between Sonatus and AWS exemplifies how generative AI can revolutionize the automotive industry. By streamlining the complexities of vehicle data management and automation, the new system makes it significantly easier for organizations to implement technical workflows. The result is a more efficient, streamlined process—achieving a remarkable reduction in policy generation time and enhancing trust through contextual accuracy.

As OEMs continue to dive deeper into the realm of software-defined vehicles, solutions like Sonatus’s Collector AI and Automator AI will play a crucial role in driving innovation and performance improvements in the industry.

About the Authors

• Giridhar Akila Dhakshinamoorthy: Senior Staff Engineer and AI/ML Tech Lead at Sonatus.
• Tanay Chowdhury: Data Scientist at AWS Generative AI Innovation Center, focused on solving business problems with generative AI.
• Parth Patwa: Data Scientist at AWS, with extensive experience in AI/ML.
• Yingwei Yu: Applied Science Manager at AWS, specializing in machine learning innovations.
• Hamed Yazdanpanah: Former Data Scientist at AWS Generative AI Innovation Center, dedicated to solving challenges using generative AI.

With such a talented team at the helm, the advancements in vehicle data management are sure to continue, paving the way for a new era of intelligent transportation solutions.