Momenta Accelerates Autonomous Driving Technology with AWS

Momenta is positioned to build an autonomous driving brain. Its core technology is deep learning-based environment perception, high-precision maps, and driving decision-making algorithms. Momenta's products include different levels of autonomous driving solutions and derived big data services. The company's strategy is to accelerate the mass production of autonomous driving technology and fully unmanned driving through data-driven algorithms and closed-loop automation.

The core requirement of an infrastructure built to support autonomous driving is the collection and processing of massive amounts of data. By using various sensors installed on-board, Momenta's test vehicles sense the surrounding environment at any time during the driving process. These sensors identify, detect, and track static and dynamic objects and to support automatic driving with the help of high-precision map data. For a test vehicle, the amount of data generated by daily road tests can reach terabytes. With all of the test vehicles combined, the amount of data that needs to be uploaded to the cloud for processing every day is massive. In addition to the test vehicles, the release of Momenta's Mpilot Highway for highway and express roads and Mpilot Parking for parking scenarios also generate a large amount of data to be processed. Therefore, a stable, reliable, high-performance, and easily expandable IT infrastructure is critical to the fast-growing Momenta.

"Momenta has been using Amazon Web Services (AWS) from the beginning for three main reasons: one is the good experience AWS has provided; the other is that AWS Activate plans provide various support in the early stage of Momenta's entrepreneurship, which has helped Momenta to start quickly and build its data service prototype. Third, Momenta started with a global development layout at the beginning of its business, and the global infrastructure coverage AWS provides has been very advantageous to that," says Momenta's research and development director Dong Weishan. 

The data collected by on-board sensors is an important foundation for autonomous driving. On the one hand, the on-board system makes driving decisions based on the information collected by the sensors at any time during the driving process. On the other hand, each test vehicle also intelligently screens and collects high-value data in real time during road testing, and the data is sent back via a 4G network or uploaded to the cloud via hard disk. Through crowdsourcing, the data is annotated and then provided to the algorithm engine for iterative training of the model, making the vehicle smarter.

Momenta uses Amazon Simple Storage Service (Amazon S3) for storage, management, and access to tens and even hundreds of petabytes of data. By using Amazon S3, Momenta can safely store any amount of data, and the data durability can reach 99.999999999 percent. "In the past three years, our data volume has grown from zero to petabytes, but Momenta's IT infrastructure has experienced no pressure," says Weishan. Combined with Amazon Elastic Container Service (Amazon ECS) for horizontal and vertical expansion, it satisfies the demand for dynamic data annotation. 

Autonomous driving requires annotating a large amount of road test data. Momenta adopts a crowdsourcing method for that data, so many people across the country make continuous requests. Amazon S3 can continuously and reliably meet such large-scale and large-scope access requests. Amazon S3 also provides a wealth of storage types. Without sacrificing performance, it performs hierarchical storage according to the value and life cycle of the data, thereby saving costs. For example, in the AWS China (Beijing) Region operated by Sinnet, Amazon S3 Glacier is used to store 1TB of data, and it costs only 3.4 Yuan ($.49) per month. Amazon S3 has built-in functions and free services, which makes it possible for a direct query of data without copying or loading data into a separate analysis platform or data warehouse. At the same time, Amazon S3 can also perform big data analysis in conjunction with Amazon EMR to provide safe driving assistance decision services.

AWS provides a wealth of IoT services to facilitate startups to build on-board applications. As Weishan says, the on-board application of autonomous driving is a very typical IoT service application scenario. On-board equipment is connected to various sensors and control systems on the vehicle. Momenta needs to build a unified management platform that can manage a large amount of vehicles and manage a large number of IoT devices.

AWS IoT Core can support billions of devices and trillions of messages. Momenta can integrate its IoT devices with cloud applications and other devices, process these messages, and route them to AWS terminal nodes and other devices safely and reliably. The application tracks and communicates with all devices at any time, even when the devices are not connected. AWS IoT Core supports multiple communication protocols such as HTTP, WebSockets, and MQTT to facilitate Momenta's ability to connect various on-board IoT devices to the AWS Cloud. AWS IoT Core performs authentication and end-to-end encryption services at all connection points to ensure vehicle management and control security. "In just two weeks, we were able to quickly build the after-market, on-board IoT prototype, complete the verification of the scheme, and open up the entire closed loop. This is also based on AWS IoT and AWS Lambda. It would have to take at least two months to develop the necessary functional components completely independently," says Weishan. Through AWS IoT Core, the alarm information sensed by the after-market on-board devices is transmitted to the cloud in time, then AWS Lambda automatically triggers according to different rules and pushes the feedback data to different B-end applications.

In the future, Momenta's on-board data mining and monitoring platform may need to manage hundreds of thousands or millions of IoT devices. With the help of AWS IoT Device Management, Momenta can easily and securely register, organize, monitor, and remotely manage IoT devices at scale, conveniently and visually manage permissions, monitor device functions, troubleshoot, inquire the status of any IoT device in the queue, and send firmware updates over the air (OTA). In addition, in terms of data transmission, AWS Snowball can migrate large-scale data to the cloud safely and efficiently at a low cost, saving operation and maintenance costs and improving developer efficiency.

Through the above IoT services provided by AWS, Momenta can easily and quickly build its vehicle cloud-integrated data platform to realize real-time position return, instruction issuance, vehicle control, and management of a large number of IoT devices. 

Artificial intelligence is the core competitive edge for Momenta, and the algorithm of its model relies on the AWS Cloud for continuous iteration. In terms of deep learning model training, Momenta sends all the valid data collected on board the vehicle back to the cloud. The cloud generates a better algorithm based on the data received, and then pushes it to and updates the vehicle so that the algorithm becomes more stable and accurate, which forms a closed loop. "Because our data volume is very large and consists of many small files, the training scale is also huge. For example, a training set is about tens of millions or hundreds of millions. Amazon Elastic Compute Cloud (Amazon EC2) P3 Instances are used to optimize the underlying and training framework to achieve support for specific scenarios," says Weishan.

Compared to the local deployment, the Amazon EC2 P3.16xlarge Instance has flexibility, scalability, is capable of automatic expansion and automatic iteration, and can achieve almost unlimited capacity while easily changing resources according to the needs of the workload. For example, in the screening process, it can flexibly configure resources and perform cloud inference according to different data volumes or project scales. In addition, strong stability and large GPU can help Momenta accelerate the speed of machine learning.

As a senior APN partner, Momenta plans to join AWS Marketplace China to extend its capabilities in artificial intelligence and machine learning to customers, and empower thousands of artificial intelligence companies based on the AWS infrastructure.

The value obtained by Momenta from using AWS is mainly reflected in the following aspects:

First, the company's innovation cycles are faster, more stable, and more efficient. By using AWS IoT services, Momenta quickly built a prototype for the mass production of autonomous driving solutions, and verified the feasibility of the design in only two weeks.

The second is to improve development efficiency and save operation and maintenance costs. Using Amazon EC2 P3 Instances enables Momenta to quickly train and tune models in a cost-effective manner. In addition, it enables Momenta to maintain a very streamlined operation and maintenance team to meet the company's needs. At the same scale, it would have required at least an operation and maintenance team of more than 10 people without the AWS Cloud.

The third is that AWS provides a variety of support. Such support includes taking note of successful cases overseas, along with market opening through the Amazon China Innovation Day, AWS technology sharing sessions, and AWS Marketplace. Momenta will expand its business to cover the global market. In doing so, Momenta can achieve rapid deployment through global AWS Regions, while having access to AWS and the global resources of Amazon to expand the market.

To learn about the IoT solutions for connecting and managing a large number of devices in the field of autonomous driving, visit AWS IoT.

To learn about the complete set of support services provided by AWS for ADAS and autonomous vehicles, visit ADAS and Autonomous Driving on AWS.