Off road, but not offline: How simulation helps advance our Waymo Driver

By | April 29, 2020
Thumbnail

COVID-19 has had a significant impact on the world, affecting people’s lives and forcing many businesses to suspend their operations. At Waymo, we’re actively monitoring the situation, taking steps to support our local communities, and contributing to COVID-19 response efforts. While Waymo has temporarily suspended its on-the-road operations as we put the health and safety of our riders, partners, and employees first, we are still driving our technology forward with our work in simulation.

Gaining 100+ years of experience in one day

Simulation is vital in the advancement of self-driving technology. At Waymo, one day in simulation is like driving more than 100 years in the real world. In simulation, we drive around 20 million miles a day, expanding the scale and complexity of our experience. True to our Alphabet heritage, our team has world-class expertise in applying cloud technology at massive scale to advance and grow our simulation efforts. To date, we have driven over 15 billion miles in simulation, and we continue to increase the velocity of our learning. In simulation, we can keep learning from each of our 20 million autonomous miles on public roads, prepare for rare edge cases, explore new ideas, validate and test new software, and continue improving our rider experience.

Amplifying real-world miles

Not every mile is created equally. A driver has to make fewer decisions while driving on a highway than driving down the bustling streets of San Francisco. In simulation, we can pick the most interesting encounters from our over 20 million autonomous miles on public roads in 25 cities across the country to maximize the efficiency of our learning.

Waymo’s simulation technology allows us to create completely simulated scenarios using data from what we’ve seen and  experienced driving in the real-world. In simulation, we can manipulate the environment around our Waymo Driver to gauge how it will respond to various situations like continuous oncoming traffic while making an unprotected turn.
Our virtual cars drive through the same scenarios that the Waymo Driver experiences in the real world but, with simulation, we build on top of this learning by modifying the scenes and evaluating possible situations. For example, what if an oncoming car is speeding while we’re making an unprotected left turn, what if a human driver behind us was distracted, or what if there was a traffic congestion at the turn? We can also manipulate the scene around us by virtually adding new agents into the situation, such as a cyclist, or increasing the speed of oncoming traffic to gauge how our Waymo Driver would have reacted. With simulation, we can amplify each scenario through a large number of variations to help assess the desired behavior of our system and then use this information to improve our Waymo Driver’s safety and performance.

Keeping simulation real

A big focus of the Waymo Simulation team is ensuring that our testing is representative of real world behavior. We have the benefit of millions of miles driven in the real world to validate and calibrate our simulation. As our software improves, the car’s position and behavior will change from the original logged scene. We ensure that as the Driver and scene evolve, we maintain realism by updating the environment around the Waymo Driver. That includes modeling agent behavior and using reactive agents–such as other cars, cyclists, pedestrians–who respond to the new position of our vehicle. We can also synthesize realistic sensor data for the car at its new position and model the scene in the updated environment.

Not only can we modify a given scenario in simulation, but we can also build entirely synthetic scenarios that we’ve never encountered in the real world, to understand how the Waymo Driver would perform. This way we continue to expand the number of miles we can simulate and validate while maintaining a high level of realism. It allows us to prepare for a range of common to rare situations and equip our software to handle them.

Before we make a change to our software, we put it through various stages of rigorous testing in simulation to assess our performance. While real-world driving is an essential part of our validation process, the majority of learning and development is done in simulation, well before the updated version of the Waymo Driver is rolled out in the real world. Through this iterative process, we can improve our software, enabling us to advance our fully self-driving technology quickly and safely.

Improving our rider experience

To improve our rider experience, we evaluate multiple comfort metrics in simulation. One approach to identify these scenarios is based on measuring the different ways people respond to the vehicle’s behavior in various driving scenarios. We use the feedback we get from our on-road testing to train machine learning models and then run them in simulation to validate how different scenarios influence rider comfort. Whether it’s figuring out the most comfortable braking speed or making sure the car drives smoothly, with simulation we continue optimizing our software for the comfort and safety of our riders.

A smooth transition to our new norm with powerful tools and infrastructure


WFH takes on a new meaning with Waymo From Home — from our newest and youngest honorary Waymonauts working on “closed course testing” to our “new remote offices.”

Developing and operating a virtual world requires large amounts of compute and, as part of the Alphabet family, we are able to use state-of-art technical infrastructure, enabling our virtual fleet to drive in simulation 24 hours a day, 7 days a week. 


While traditionally, most of the tooling used for simulation requires engineers to be on powerful machines at their work stations, at Waymo, we build and use many technologies that enable more flexibility and fast development from anywhere. For example, having web versions of the most commonly used applications has allowed us to ensure that our engineers are productive while working from home and that we continue driving our technology forward in these unprecedented times. This software provides secured remote access to our engineers, loads quickly, boasts greater functionality and grants us more flexibility.

Even though we’re not currently driving in the real world, we’re continuing to gain thousands of years of experience through simulation during this time. While we can’t wait to get back on the road and start serving our Waymo One riders again, our Waymo Driver continues getting safer and smarter every second until then. 

We hope you are staying safe and healthy and look forward to seeing you on the road soon!

Join our team and help us build the World’s Most Experienced Driver™. Waymo is looking for talented software and hardware engineers, researchers, and out-of-the-box thinkers to help us tackle real-world problems, and make the roads safer for everyone. Come work with other passionate engineers and world-class researchers on novel and difficult problems—learn more at waymo.com/joinus.


Read the original article

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.