Guruswamy Ganesh of Western Digital (which acquired Sandisk) kicked off the meeting and introduced their newly launched automotive interests focusing on the safety, quality, reliability, and endurance that will be required as we move towards self-driving, autonomous vehicles. It’s a given that the number of microprocessors is destined to increase as cars drive themselves - and an increase in stored data goes hand in hand: predicted at 1 terabyte per vehicle by 2025. The concept of the car being an IoT device is a certainty, and connected cars will consequently produce a huge amount of data, some for local storage, some uploaded to the Cloud. Intelligence at the edge is going to play a key role in growth of automotive market segment, compensating for the limitations of cloud networking (intermittent connectivity, cost, variable datarates). This is an exciting time to be part of the automotive ecosystem, to be knee deep in the innovation as it happens.
Speaking to the progress and opportunities mapping offers the market, Eric Gunderson of MapBox donned the role of analyst looking at the growth from the 2004 “print and carry along a Mapquest map” solution, through to the current interactive, GPS, crowdsourced mapping we use today. Live maps have become fundamentally central to how we drive, with Google reinforcing that value with its claim that “we have to spend billions of dollars…” on maps “…and it’s worth it!” Data from Goldman Sachs was shown that estimated revenues for mapping data will rise as high as $9.2B per annum in short order. Gunderson predicts open mapping solutions will capture a share of that bounty, breaking the false connection between "open" and "free". Open map layers will allow collaboration to maximize the value of that layer of data, but still allow plenty of room for differentiation in other layers, or in algorithms and AI.
Our Fireside Chat brought Arnold Meijer of TomTom Maps and Hossam Bahlool of TeleNav to the stage to explore the benefits and challenges of HD maps and how that map data might be leveraged. The point was made that HD map data has TWO consumers in the autonomous car: 1) the AI that can do some self-driving and 2) The human. Both "customers" have very different capacities to absorb data, and different display needs. For the robot, it can handle immense quantities of data for the purpose of safe driving, while the human passenger may see map data as "infotainment", like a passenger on a commercial airline. Both speakers' companies evolved from turn-by-turn nav into the HD maps space. The main challenges facing the transition from SD to HD were seen as follows:
- Cost – HD mapping is significantly more expensive, from the cost of updating the data to the cost of connectivity.
- Connectivity – bandwidth consumption will be the major differentiator for mapping solutions, with companies needing to consider caching, utilizing wifi, regional updating (tiling).
- Consumer expectations – with on-street parking given as an example where it’s not just availability but appropriateness has to be considered.
- Real Time data – where even the most granular HD map data can quickly become stale-dated, various methods of keeping it fresh need to be deployed, such as crowdsourcing, open sourcing, and ongoing collection
Exploring the path to Autonomony, Andy Parsons of Increment P, Colm Lysaght of Western Digital, Jeff Wuendry of Velodyne LiDAR and Kevin Tsurutome of Comtech Telecommuncations, joined the panel and were asked to choose … Maps or Sensors? It was quickly established that one cannot function without the other, even looking forward to a future where sensors become inexpensive and ubiquitous. The vehicle is always going to need to know where it is going (HD Maps), what is around it at any given time (maps and sensors), exactly where it is (sensors), and for safety, what's going on around it right now (sensors). Building on some of the discussion points from the fireside chat, the panel agreed that maps are essential as the foundation for safe, efficient navigation. They need to be reliable, regularly updated, and futureproof. Sensors would augment the experience, to manage the unpredictable, be it weather conditions, road conditions, accidents or congestion. With LiDAR being acknowledged as the “gold standard” of sensors, the panel discussed the expense associated with using sensors which may be acceptable to mapping companies and ride sharing solutions who earn money with the vehicle, but prohibitive to the current consumer because of the higher CapEx. The conversation moved on to the data demands of autonomous driving and the panel agreed that in the early days, we are likely to overestimate requirements. There will clearly be an incredible amount of data required, not only for navigation but also to augment V2X communications, telematics and maintenance, autonomous driving algorithms, AI and machine learning. Data will need to be managed between local storage and connectivity, via caching, least-cost routing, and wifi uploads, to ensure critical data can be filtered and compressed to ensure uninterrupted data transmission when it is most needed. The panel posed that, with a robust mapping backbone, 10-20 kilobytes per kilometer might be sufficient bandwidth to accommodate the on-demand, changing layer.
It was clear from the conversations during the networking break that the panel had provoked serious thought. Following the break, we brought 7 startups to rapid fire:
Metrotech – providing a neutral, trusted, third-party digital streets platform.
Katla Labs – introducing “SideKick” an all-in-one driving assistant.
MagicGPS – leveraging wi-fi while roaming.
DeepMap – offering HDMaaS – HD Maps as a Service for full autonomous vehicles.
RFNav – developing all weather maps for autonomous vehicles.
Civil Maps – creating a new generation of maps facilitating cognition for cars.
Helm.ai – proposing safe algorithmic navigation for autonomous vehicles.
Thanks, once again to our hosts, and to all our speakers. Members may find all the meeting’s presentations uploaded in the member library.