Engineering

The Future of AI in Self-Driving: Facilitating and Benefitting from Scale

Sterling Anderson — Thu, 12 Sep 2024 10:00:00 GMT

We founded Aurora nearly eight years ago to deliver the benefits of self-driving technology safely, quickly, and broadly. This mission has given purpose to our team and direction to our development: our product and our business are grounded in improving safety, creating value quickly, and scaling broadly.

Paving the Road for Aurora’s Growth

Chris Urmson — Sat, 03 Aug 2024 22:10:02 GMT

Last week we raised total gross proceeds of $483 million, which adds to our $1 billion of liquidity as of the end of June. With runway well into 2026, we expect this incremental capital to fund the initial phase of our scaling strategy. I feel energized by this momentum as we prepare to launch driverless trucks in Texas, planned for the end of this year.

AI Alignment: Ensuring the Aurora Driver is Safe and Human-Like

Drew Bagnell — Fri, 26 Jul 2024 15:55:20 GMT

Aurora believes the future of self-driving depends on the expansive use of the power of AI combined with rigorous validation and verification. At a time when researchers, industry, and the broader public are generating more questions and curiosity about the potential of AI, we’re taking you under the hood of Aurora’s approach in a series of blog posts.

Previously, Chris unpacked the concept and application of Verifiable AI, while I weighed in on how we build transparent AI systems. Today, I'll talk through how we ensure alignment in AI.

AI Transparency: The Why and How

Drew Bagnell — Wed, 17 Jul 2024 05:51:32 GMT

Increasingly, researchers, industry, and the public are taking AI seriously as the heart of self-driving. It's been gratifying to see Aurora's pioneering approach to validated AI in self-driving become popular across industry and research. When we founded Aurora, we knew that the future of self-driving would depend on expansive use of the power of AI combined with rigorous validation and verification.

Aurora’s Verifiable AI Approach to Self-Driving

Chris Urmson — Thu, 13 Jun 2024 15:30:00 GMT

Artificial Intelligence (AI) is essential to the success of any modern complex system, including self-driving. We’ve been saying this for 7 years. There is simply no other way to get human-like driving proficiency or handle the diversity of conditions that are experienced on our roads today.

Continuous real-time sensor recalibration: A long-range perception game-changer

the Long-Range Perception Team — Wed, 15 Mar 2023 17:00:00 GMT

To safely power autonomous semitrucks at high speeds on highways, the Aurora Driver must accurately and reliably perceive objects and vehicles at great distances. To do so, the Aurora Driver is equipped with sensor pods made up of ultra-high-definition cameras, imaging radar, and our proprietary FirstLight lidar. Data from each of these sensors must be processed and fused into a holistic and dynamic picture of the truck’s environment.

Introducing Au: Our open source C++ units library

Chip Hogg — Wed, 08 Feb 2023 18:00:00 GMT

Units are mission-critical

When software controls real objects in the physical world, it's critical to get the units right for every variable. This was starkly illustrated by the now-infamous demise of the Mars Climate Orbiter in September 1999. The spacecraft reported its impulse measurements in US customary units as pound-force-seconds, but the trajectory calculator interpreted them in metric (SI) units as newton-seconds—more than four times smaller. Ultimately, this mismeasured steering led to the total loss of the spacecraft, wasting hundreds of millions of dollars.

Aurora’s Data Engine: How we accelerate machine learning model workflows

Ankit Aggarwal and Vinay Anantharaman — Tue, 17 Jan 2023 18:00:00 GMT

Machine learning (ML) models form the backbone of autonomous vehicle (AV) development. Everything from perception to motion planning is powered in part by the datasets and models developed in a cycle commonly referred to as the “Data Engine.” The Data Engine lifecycle starts with identifying the type of data required to support or improve an AV capability—for example, to detect emergency vehicles, our perception system needs lots of sensor data of emergency vehicles in different situations. This is followed by iteratively mining and labeling the data to turn it into a usable dataset for ML model training. Once a model is trained, it then goes through a number of sub-system and system-level evaluations and those results are fed back to the Data Engine for the next iteration. Our ability to quickly and safely develop, deploy, and continuously improve our self-driving technology is largely dependent upon the efficiency and speed of this lifecycle.

Lightbox: Autonomy visualization at Aurora | Aurora

the Lightbox team — Thu, 15 Sep 2022 04:00:00 GMT

How we built a new tool that lets autonomy engineers move 10x faster

To build autonomous trucking and ride-hailing products, we need to understand what the autonomy system is experiencing. In other words, we need to be able to see what the Aurora Driver sees—is it picking up those construction cones, is it registering that speed change sign, does it notice that motorcycle coming from behind? What’s more, we need to be able to see how and why the Aurora Driver sees what it sees—are some of the cones obstructed and not perceptible, is the sign confusing or illegible, is the motorcycle still recognizable when it suddenly does a wheelie?

Access to all of this information allows our operations teams to quickly spot and diagnose issues that appear while our vehicles are out testing; it allows our simulation teams to reliably test our technology offline at scale; and it allows our engineering teams to understand how the code they write affects the way our vehicles perform out in the real world. Engineers on Aurora’s Autonomy team, for example, generally run thousands of short simulations to evaluate whether a code change or new model should be implemented. Small adjustments often result in a number of improvements and regressions, and autonomy engineers typically have to inspect dozens of these short video snippets to verify whether those improvements and regressions were expected or whether they warrant deeper investigation.

We’ve built several custom tools over the years that enable autonomy visualization for various key workflows, but as we’ve grown and more people have begun using these tools, we’ve run into a few challenges:

Our autonomy visualization tools were built on different tech stacks at different times.
Each tool was built for a specific use case and has trouble generalizing to new use cases.
The tools are slow and struggle to keep up with increasing demand.

To solve these pain points, a cross-functional group of folks from product, design, and engineering got together and decided to see if we could develop a new autonomy visualization platform. After conducting rigorous user research, engineers on Aurora’s Visualization team began experimenting with a first-principles approach and our product and design team began mocking up visual prototypes. Several iterations later, we landed on a new visualization application that would allow our autonomy engineers to move 10x faster.

A piece of the perception puzzle: Harnessing the range view | Aurora

Shivam Gautam — Wed, 31 Aug 2022 04:00:00 GMT

At Aurora, we pride ourselves on building state-of-the-art machine learning approaches for autonomy. We are continuously innovating in lidar sensor development, allowing our autonomous vehicles (AVs) to see lidar points from farther away, faster, and with greater precision than ever before. A publication by several Aurorans exploring a novel machine learning approach to processing lidar data, “LaserFlow: Efficient and Probabilistic Object Detection and Motion Forecasting,” (Gregory P. Meyer, Jake Charland, Shreyash Pandey, Ankit Laddha, Shivam Gautam, Carlos Vallespi-Gonzalez, Carl K. Wellington) was recently acknowledged for winning a 2021 Best Paper in Robotics and Automation Letters at IEEE’s 2022 International Conference on Robotics and Automation.

Batch API Part 3: Supercomputing at Scale | Aurora

the Compute Infrastructure team — Mon, 04 Apr 2022 04:00:00 GMT

As Aurora continues to grow and prepare for the launch of Aurora Horizon and Aurora Connect—our autonomous trucking and ride-hailing products—reliable and scalable computing power only becomes more important.

Batch API is designed to efficiently and flexibly manage the massive amounts of computing power we require, allowing our teams to run millions of tasks every day. But to maintain optimal performance, the owners of these tasks also need to be able to analyze the effectiveness of all of this computing power. And to do that, they need quick and reliable access to Batch API data.

In the third and final part of our series on Aurora’s Supercomputer, Batch API, we will discuss the challenges of logging user data and debugging on a massive scale, and how we provide users with tools to evaluate the performance of their jobs via access to historical analytics data. We’ll also talk about the future of our platform and how Batch API is scaling to a billion tasks a day.

System analytics and debugging

TetrysTables: Speeding up machine learning model development with shared datasets

Yevgeni Litvin — Mon, 14 Mar 2022 04:00:00 GMT

Autonomous vehicles understand the world by processing enormous amounts of data acquired via sensors. We use this data to train machine learning (ML) models to detect objects surrounding the vehicles, forecast their future behaviors, and plan the autonomous vehicles’ next movements.

Batch API Part 2: Resource Management | Aurora

the Compute Infrastructure team — Tue, 25 Jan 2022 05:00:00 GMT

Computing power is an extremely valuable resource that can determine a technology’s ability to scale. We believe self-driving technology will someday soon be ubiquitous, and to deploy Aurora’s products broadly, we’ve invested heavily in the design of a flexible and efficient computing infrastructure that scales with our business.

Batch API Part 1: Aurora’s Supercomputer | Aurora

the Compute Infrastructure team — Thu, 09 Dec 2021 05:00:00 GMT

Self-driving is full of technical challenges, and as a company focused on delivering the benefits of self-driving technology safely, quickly, and broadly, Aurora is solving these challenges using a variety of disciplines such as machine learning, computer vision, simulation, and vehicle log ingestion. Much of the work we do is extremely resource-intensive in terms of computing power. On any given day, we process 5-12 million tasks through our computing infrastructure, ranging from simulations, vehicle log processing, high-resolution map building, perception inference, calibrations, and more.

Perception at Aurora: No Measurement Left Behind | Aurora

the Aurora Team — Sun, 14 Nov 2021 05:00:00 GMT

How we’re blending state-of-the-art machine learning and engineering in our perception system to help the Aurora Driver handle the unexpected

Imagine you’re driving on the highway when something tumbles out of a flatbed truck in front of you. You can’t quite see what it is, but you can already tell it might end up in the middle of your lane. What would you do?

Forecasting: Understanding Interaction | Aurora

Drew Bagnell — Wed, 10 Nov 2021 05:00:00 GMT

It's tough to make predictions, especially about the future.

The Technological Foundation Accelerating Development of the Aurora Driver | Aurora

the Aurora Team — Thu, 14 Oct 2021 04:00:00 GMT

At Aurora, we recognized early on that upfront, foundational investments in key technologies would enable us to build a safe and scalable Aurora Driver. Aurora Illuminated was a showcase of this approach—what we call self-driving 2.0.

Scaling Simulation | Aurora

Jessie Smith and Viktor Lundqvist — Thu, 11 Feb 2021 05:00:00 GMT

At Aurora, we build the right tools for the right problems. Here’s how we’re auto-generating the hundreds of thousands of virtual tests needed to rapidly develop and deploy the Aurora Driver.

Aurora’s Software Engineering Tenets | Aurora

the Aurora Team — Mon, 23 Nov 2020 05:00:00 GMT

In software engineering, we're reaping the benefits of early investments in machine learning infrastructure, getting fast, detailed feedback from cutting-edge simulation tests, and building out key teams from Site Reliability Engineering to Developer Infrastructure. And this is still just the beginning. As we continue to grow, we’re thinking deeply about how we can support the thousands of engineers we’ll hire down the road.

As many of us in the industry know, scaling gracefully is a balancing act. We keep our standards high by staying away from demo-driven development, following style guides, and writing code that other engineers can understand and maintain. It’s equally important to actively prevent gatekeeping and bureaucracy. We make a point of hiring experts, and we win when we allow them to creatively apply their knowledge throughout the stack.

So, to help us do more of what works (and less of what doesn’t), we’ve collectively developed five Software Engineering Tenets that guide us day-to-day. Below, we share how they enable innovation, help us make good decisions, and generally make being an engineer at Aurora awesome.

Aurora's Online to Offline: Turning real-world experience into virtual tests | Aurora

the Aurora Team — Tue, 12 May 2020 04:00:00 GMT

On-road events become more impactful when we add them to our Virtual Testing Suite and use them to continually improve the Aurora Driver.

At Aurora, we’ve adopted a “smarter, not farther” approach to on-road testing. That is, instead of blindly pushing to drive more and more miles, we’ve continued to focus on collecting quality real-world data and on getting the most value out of every data point. For example, we amplify the impact of real-world experience by flagging interesting or novel events and incorporating them into our Virtual Testing Suite.

FMCW lidar: The self-driving game-changer | Aurora

the Aurora Team — Thu, 09 Apr 2020 04:00:00 GMT

Virtual Testing: The Invisible Accelerator | Aurora

the Aurora Team — Fri, 17 Jan 2020 05:00:00 GMT

On-road testing gets the spotlight in the industry, but there’s a lot of meaningful work happening behind the scenes at Aurora to deliver self-driving technology safely, quickly, and broadly. One major accelerator is our Virtual Testing Suite, which allows us to conduct millions of valuable off-road tests per day.

Since Aurora’s inception, we’ve spoken at length about the many ways we’re fueling the rockets for success. This has included our design for accurate, scalable HD maps, our approach to creating robust hardware from first principles, and our acquisition of Blackmore FMCW lidar. Over the past year, we’ve made tremendous progress in every facet of the Aurora Driver, allowing it to master increasingly complex maneuvers in equally complex environments.

Teleassist: How Humans Collaborate with the Aurora Driver | Aurora

the Aurora Team — Mon, 16 Dec 2019 05:00:00 GMT

Our system leverages expert human guidance to improve the efficiency and reliability of the Aurora Driver.

Self-driving vehicles will save lives, improve access to transportation, materially change the economics of goods delivery, and allow cities to reclaim the space currently occupied by parking lots. We’ve spoken at length about our development approach and the myriad of ways we’re designing our Driver with safety and efficiency in mind. But to be truly impactful, our self-driving fleet must also operate reliably at scale.

Safety and reliability are distinct objectives. Safety sets a hard prerequisite for deployment: we will keep trained personnel behind the wheel until we are confident that our Driver can operate more safely than the average human. Reliability affects the quality of our service after we make that determination and deploy self-driving vehicles without safety drivers.

We will not deploy our vehicles in a commercial setting until we are confident that the Aurora Driver meets our safety standards. To facilitate deployments in new operational domains while maintaining high reliability, Aurora is investing in teleassist. This set of technologies allows trained specialists to monitor our vehicles and provide high-level guidance when needed.

Read on to learn more about teleassist at Aurora, how it’s different from remote control, and what this investment means for our business moving forward.

Product Design from First Principles and Experience: Hardware at Aurora | Aurora

the Aurora Team — Tue, 05 Nov 2019 05:00:00 GMT

Nearly three years ago, Aurora was founded with one mission: to deliver the benefits of self-driving safely, quickly, and broadly. From the earliest days of the company, we knew that achieving this mission would require a deeply experienced team and a meticulously-designed Aurora Driver consisting of the hardware, software, and data services required to navigate vehicles safely through the world.

The Atlas: Our HD Mapping System | Aurora

the Aurora Team — Thu, 17 Oct 2019 04:00:00 GMT

How we design the lightweight, scalable maps that serve as a foundation for the Aurora Driver.

Most of us are better drivers in places we’ve driven before. We know we should drive cautiously in areas where there might be cyclists, we learn when to yield in complex intersections, and we avoid stress (and honking) by changing lanes early out of an exit ramp we know is coming up. Self-driving vehicles also benefit from foreknowledge about their route, which they get from specially designed high-definition (HD) maps. These maps are extremely detailed and contain far more information than those used in typical in-car navigation systems.

We’ve been developing our own HD map for the Aurora Driver called the Atlas. The Atlas contains information about static scenery and road infrastructure, making our Driver more efficient by reducing the amount of data it has to process in real-time. For example, we build stop signs into the Atlas so our Driver always knows they’re coming, even when they’re faded or hidden behind a parked car.

Without thoughtful design, HD maps can become a morass of dense, heavyweight data that requires significant time and resources to maintain. At Aurora, we view three pieces as critical for maps that enable the Aurora Driver to be deployed safely, quickly, and broadly: (1) the right content, at (2) the required accuracy, and (3) the design, data, processes, and tools that enable rapid updates at scale. We’ve optimized every part of the Atlas’s design with these goals in mind.

Read on to learn more about our approach to developing this foundational piece of the Aurora Driver.

The Offline Executor: Virtual Testing and the Aurora Driver | Aurora

the Aurora Team — Tue, 20 Aug 2019 04:00:00 GMT

Our approach to simulation reflects our design philosophy—we invest time to create tools expressly for self-driving cars, increasing operational safety down the road.

There’s a lot of talk in our industry about how many miles companies have driven on the road. While there is value in real-world miles, we believe the safest and most efficient way to deliver self-driving technology is by incorporating strong virtual testing capabilities into our development.

At Aurora, we invest time in building smart tools and processes that enable us to quickly accelerate development, all while keeping safety at the forefront. Our simulation work is no exception. Unlike many in the industry, we’ve taken the time to develop our own solution to analyze system performance through simulated autonomous driving scenarios, rather than rely on game engines or other pre-built software. We call this solution the offline executor.