Addressing the data problem with autonomous vehicles

Customer:
Multinational Automotive CompanyChallenge:
- Reduce iteration cycles and other processes to accelerate the development of the autonomous vehicle fleet
- Manage the data generated by self-driving vehicles and identify novel situations to increase consumer safety
- Accelerate the process of “teaching” vehicles how to handle unique environmental conditions
Solution:
- Expedite data analysis to reduce the learning curve for smart-car AI controls
- Leverage DXC Robotic Drive to manage massive data flows in native vehicle data formats
- Automate deployment of functional testing
Results:
- Reduced “time to drive” and “time to analyze” to accelerate delivery of autonomous vehicles to the marketplace
- Automated end-to-end approach from data ingestion and processing, via neural network training to functional testing, and in-car deployment
- Accelerated progression through the sequences of autonomous driving levels for increased ROI
The automotive industry has entered a new period of innovation focused on delivery of self-driving vehicles, with established auto manufacturers and tech-savvy outsiders alike pouring billions of dollars into the development of autonomous vehicles to carry people and transport freight.
This new cycle of invention and experimentation is being driven by changing mobility and customer demands, as well as safety and environmental concerns. But it is made possible by recent strides in IT — the ability to capture massive amounts of data and apply a deep learning algorithm to images to recognize other vehicles. DXC Technology was hired by a global auto manufacturing company to help more efficiently capture the data being gathered by its self-driving vehicles and use that information to advance development efforts.The autonomous vehicle scaling challenge
Innovation was once an in-house exercise, but as OEMs develop autonomous driving algorithms on their own, they are now collaborating with other automakers. Like other companies in the industry, this global auto manufacturer was challenged with scaling the development of autonomous vehicles, which requires a wide range of skills and capabilities. In many cases, these have less to do with how to make a vehicle go and everything to do with how to make it safe and smart, such as expertise in artificial intelligence (AI), machine learning, next-generation computer science and data management.
It is imperative that an autonomous vehicle be able to “see” where it is going, detect and avoid hazards, and transport passengers safely with little or no human input. This requires collecting and processing massive amounts of data. Real-time data about the environment — weather, road conditions, other vehicles, pedestrians and street signs — combined with information about the vehicle and the intelligence needed to make instant driving decisions, generates up to 4 terabytes of data per hour for one test vehicle. That data is analyzed for real-time events as managed by the vehicle, but later the data is analyzed by manufacturers, scene by scene, to identify novel driving conditions that can be used to inform the AI that underlies the vehicle’s autonomous operations. The problem of scale becomes apparent as hundreds of hours of recorded data must be searched for unique instances that can then be used to “train” the autonomous driving algorithms on how to manage those new situations. This can take weeks of research.Enabling autonomous acceleration
In the highly competitive auto industry, it is essential to have the ability to process and act on autonomous driving data quickly and efficiently. This global auto manufacturer was looking to reduce iteration cycles and other processes to accelerate the development of its autonomous vehicle fleet. A team of DXC technologists helped the company speed up its self-driving car R&D, and has been recognized by DXC with a 2019 Technical Excellence award for its work.
The DXC team, working with the DXC Robotic Drive, autonomous driving platform, toolkit and accelerators, built a solution to collect and manage the massive data streams created by the auto manufacturer’s test fleet. AI and deep learning solutions were created to analyze the data quickly and to automatically flag interesting encounters that could provide valuable lessons to the fleet’s autonomous driving software.
Built from standard components on an open source ecosystem, DXC Robotic Drive helps automakers rapidly analyze data in the format that was recorded by the vehicle, a major time-saving step, versus having to convert the data. DXC Robotic Drive also provides auto manufacturers with large containerized compute clusters and provides the software infrastructure for orchestrating Deep Neural Network training. What’s more, DXC Robotic Drive gives autonomous driving R&D teams a platform to manage and search the data collected by a test fleet in an urban driving setting— a capability that’s needed to move autonomous vehicles from the experimental stage to the showroom floor.Speeding up data ingestion
Successful application of the DXC Robotic Drive solution’s autonomous driving platform, toolkit, accelerators and expertise helped produce numerous benefits for the auto manufacturer. This included faster data ingest rates (minutes rather than days), faster development of algorithms, shorter iteration cycles, a 50 percent reduction in time to drive and significant algorithmic performance gains.
Ultimately, the DXC team’s efforts will help the customer’s vehicles reduce system disengagement rates (when driving control is turned over to a human), which today are at a near-human level. This is a critical step forward that helps DXC’s auto manufacturing customer build and manage a fleet of Level 4 autonomous cars for the busy streets of a major US city and will eventually be capable of fully autonomous operation.