Delivering a fully autonomous driving (AD) vehicle remains a key priority for both manufacturers and technology firms (“firms”). However, passenger safety is now a top-of-mind concern due in great part, to fatalities resulting from driving tests over the past years. Firms are now emphasizing:
• New test software features that better detect, monitor, and respond to both driver and road conditions.
• New safety rules governing test vehicle operations. For instance, test vehicles in Pittsburg that are managed by Uber, will now operate only on weekdays in daylight with two (2) trained employees seated within each vehicle at all times.
This greater focus on passenger safety is echoed by the U.S. Department of Transportation, which has stated that AD future relies on the development of vehicle technology and the appropriate operating rules and response protocols. With this in mind, developing an AD vehicle has become more expensive and will take longer. Developers will have to deal with:
• More data. Developing more accurate inference models for AD will require more testing and therefore more training data. The data ingested and managed to create a more accurate inference model for how to drive under multiple scenarios is massive. One test vehicle alone can produce up to 15 Terabytes of data per hour. Collecting, pooling, curating and managing all that data has been and will remain a key challenge for developers.
• Data sharing and privacy. Potentially more mergers, acquisitions, and collaborations amongst AD industry participants. By one estimate, there were 53 LiDar companies alone in California. Sharing and managing multiple party data with their various policies has been and will remain a key challenge for developers. Now add in data privacy rules like the General Data Protection Regulation (GDPR), ugh.
Pooling this data together will require high-speed transport with massive throughput, large scale storage capacity on hybrid cloud, rapid meta-data tagging, and intelligent archiving as this data ages. Data policies for sharing will need to be clearly outlined and enforced for one or many companies involved in the development. In response, IBM has invested in technologies to address the storage, workflow and management system for AD; they include:
• Blockchain which offers a way to enforce who can share what data.
• High speed data transfer for hybrid cloud.
• Hybrid multi-cloud storage including all-flash arrays, cloud object storage, file/block storage, and cost-efficient tape
• Meta data management
• Auto labeling for video
• Workflow and IT resource management
• Reference architectures for accelerating the ingest, preparation, and use of data in deep learning and training of AI models. The IBM Systems Reference Architecture for AI and the newly released IBM Spectrum Storage for AI can help drive AI development productivity and streamline AI data pipelines.
IBM Spectrum Storage for AI simplifies deployment, provides performance, and enables extended data management for deep learning and training of inference models. IBM Spectrum Storage for AI, combines IBM Spectrum Scale, Power Systems and NVIDIA GPUs, to enable large scale, faster training. Watch this 15 minute video and gain visibility into how your storage choices can impact your AD/ADAS strategy. http://ibm.biz/Bdz2sa
Learn more about these and other IBM solutions for autonomous driving (AD) and advanced driver automation systems (ADAS) at https://www.ibm.com/it-infrastructure/storage/ai-infrastructure