GCP – Nuro drives autonomous innovation with AlloyDB for PostgreSQL

Editor’s note: Nuro, a robotics company that develops technology for self-driving vehicles, needed a data platform that could handle complex data processes and support continuous AI model improvement. By migrating to AlloyDB for PostgreSQL, Nuro gained the scalability, high performance, and advanced query capabilities needed to power AI-driven insights across millions of data points while reducing operating costs. AlloyDB AI further enables Nuro to perform complex similarity searches on vector embeddings, supporting continuous improvement.

Nuro’s mission is to make daily life better through robotics. 

One of the ways we achieve this is with Nuro Driver, an AI-powered technology that automakers and mobility providers use to develop autonomous vehicles for personal use, delivery services, and ride-sharing applications. 

Naturally, creating self-driving technology that’s truly safe and reliable takes more than just innovation — it requires a platform capable of processing vast amounts of data and adapting to continuous learning cycles. That’s why we needed data infrastructure that could handle our growing volumes of complex data and support essential processes like data discovery, labeling, and rapid evaluation. 

As we navigated options for a managed SQL database that could handle these challenges and build on our existing PostgreSQL setup, we explored several options. We ultimately arrived at AlloyDB, a high-performance, fully managed PostgreSQL-compatible database on Google Cloud, for its superior performance, ease of use, and seamless integration.

Gearing up for autonomous data growth

Transitioning to a new data infrastructure can often be disruptive, but with AlloyDB, the process was seamless. The migration from our existing PostgreSQL environment required zero downtime and one-click setup. This allowed for continuous fleet operations without interruptions to deliveries or model training. AlloyDB now powers our core transactional and analytical workloads, managing crucial metadata for logs, trips, simulations, and real-time autonomy issues.

Operating across multiple cities, we rely on Google Cloud’s global availability to collect and manage petabytes of data for AI model training, evaluation, and simulation — with quick turn-around. This infrastructure enables analysis for refining route optimization to find challenging scenarios so our AI models can learn based on real-world on-road performance. AlloyDB plays a critical role in this ecosystem, efficiently processing large query volumes while supporting the rapid, data-driven decisions essential to autonomous operations.

Beyond performance, AlloyDB’s fully managed service reduced the burden of scaling and maintenance, allowing our team to focus on improving AI models rather than database administration. Its advanced query capabilities and deep integration with Google Cloud streamlined workflows, helping us iterate on autonomy models faster. With improved efficiency and reliability, our fleet can continuously evolve.

A data platform built for the long road ahead

Google Cloud is always innovating new ways to advance autonomous driving. We recently migrated all our vector embeddings to AlloyDB AI, enabling ML-based similarity searches across millions—and sometimes hundreds of millions—of vectors. With AlloyDB’s vector store and advanced indexing using ScaNN, our autonomy team can run complex similarity searches that quickly identify scenarios where Nuro Driver can learn and improve. AlloyDB’s high query performance for both transactional and analytical tasks ensures we can scale our dataset continuously, allowing us to train models on increasingly complex road conditions without performance bottlenecks.

To support these capabilities and improve performance, we’ve built a comprehensive ecosystem on Google Cloud. Cloud Storage serves as our primary storage for autonomy logs, on-road operation data, simulation records, and ML evaluation files. Using change data capture from Datastream, we replicate AlloyDB data to BigQuery in near real-time. This creates a unified flow that supports business dashboards and provides detailed, real-time analytics on autonomy performance. BigQuery serves as the main backend for analytical metrics, enabling precise evaluation and validation of the Nuro Driver.

Additionally, we use Spanner for storing log namespace metadata, while Firestore, Datastream, and Memorystore support various other applications, making our data management flexible and efficient. This diverse set of databases on a single cloud platform not only centralizes data management but also enables real-time insights and seamless data access. It’s the robust, scalable foundation we need to drive reliable autonomy at scale.

AlloyDB takes the driver’s seat in Nuro’s data transformation

Since migrating to AlloyDB AI, we’ve seen a substantial reduction in the operational costs of storing and searching embeddings. AlloyDB AI’s horizontal scalability has proven to be the most cost-effective solution for our needs, allowing us to add several new types of embeddings across applications without concerns over performance. With ScaNN indexing, our searches now yield over 20,000 high-precision results in seconds, outperforming alternative indexing methods like IVF and HNSW in both quality and scalability.

Our partnership with Google Cloud has also been invaluable. We have continuous access to innovations from the Google Cloud team, and we can easily meet any database requirement by leveraging their extensive suite of products. This support has accelerated our development, enabling us to focus on what matters most — advancing autonomous technology.

Looking forward, Google Cloud remains our primary cloud platform. Relying on its global presence and infrastructure, we can expand our services to new customers worldwide, all while maintaining the high standards of reliability and performance our team depends on. Google Cloud gives us the green light to tackle future challenges in autonomous driving, remove potential roadblocks, and keep innovation on the fast track.

Ready to get started with AlloyDB in your own environment? Check out the following resources:

• Discover how AlloyDB combines the best of PostgreSQL with the power of Google Cloud in our latest e-book.

• Try AlloyDB at no cost for 30 days with AlloyDB free trial clusters!

• Learn more about AlloyDB for PostgreSQL.