DeepMap customers include Daimler, Einride, Ford, and Honda. Investors include Accel, Andreessen Horowitz, and NVIDIA. How has DeepMap established such a strong presence in just four years?
To learn more, CIOReview met with DeepMap’s VP of Engineering, Andy Shaw. Shaw joined DeepMap in 2019, following years of experience leading engineering teams at VMware and Big Switch. CIOReview discussed a wide range of topics with Shaw, from artificial intelligence and HD mapping to leading a team during a global pandemic.
High-Definition Mapping: Essential for Autonomy
Shaw contends that HD mapping is mission-critical for safe self-driving vehicles. He explains that an HD map contains a description of the world in terms that a self-driving car can understand. It provides all of the information about a road’s geometry as well as the semantic rules and associations that are used to ensure the car obeys traffic rules and conventions. “We offer solutions that can be customized for any high-definition map requirement, on any sensor configuration, for every level of autonomy,” says Shaw.
The Role of AI at DeepMap
At DeepMap, says Shaw, developers utilize sophisticated AI tools and techniques to build HD mapping technology from Level 2+ to Level 4. Examples of AI at DeepMap include:
• Deep learning is used to automatically detect and create 3D map features/landmarks (signs, signals, lane lines, etc.) from input sensors.
• Each point in the DeepMap 3D occupancy map carries a semantic label populated from perception results. This helps to determine objects which should not be included on a map, such as people and vehicles.
• The DeepMap API supports real-time in-car perception for detecting changes in the map—for instance, cones for road construction, changes in lane lines, or new signs or removed signs.
Shaw points out that earlier this year, DeepMap was recognized as a “Best Machine Learning Startup to Work for” in 2020, based on an analysis of Glassdoor data.
Every map provider will claim that they have coverage, accuracy, and completeness in their data, says Shaw. Taking a real-time system centric approach and making a map that is deployable at a large scale is where DeepMap’s approach differentiates. “We’ve sought a deeper understanding of the specific and varying roles the map plays as a component in autonomous systems and tailored ours accordingly.
This means building in the flexibility to meet customers’ different needs in an economical way.”
We think about mapping as a complete service that optimizes the entire process - from map creation to map consumption and map serving. To ensure map quality and consumption performance, maps need to be made by self-driving cars and for self-driving cars, creating a virtuous cycle
He adds: “We think about mapping as a complete service that optimizes the entire process -- from map creation to map consumption and map serving. To ensure map quality and consumption performance, maps need to be made by self-driving cars and for self-driving cars, creating a virtuous cycle.”
Leadership during the Pandemic
Shaw says, “At the core of our company is a team-oriented culture.” During the COVID-19 global crisis, the company’s top priority is to create an environment in which the team can be safe and productive. Shaw says the company’s productivity metrics are being maintained, and in some cases improved, during the work-from-home period.
With fewer vehicles on the road, data collection is becoming more efficient as the company continues to map highways in California as well as other U.S. and global locations, while observing safety and social distancing guidelines. Shaw adds: “During the pandemic, we’ve seen a growing interest in mapping for contactless delivery. We recently began providing mapping services to JD, a large retailer in China, for an autonomous delivery initiative.”
The DeepMap Roadmap
According to Shaw, the company originally focused on developing mapping technology for Level 4-5 driving, which includes self-driving robo-taxis, but recently has seen market demand for ADAS (Advanced Driver-Assistance Systems) and Level 2+ solutions.
He says: “Moving forward, it is clear that technologies such as automation, AI, and machine learning will control the path of automotive development. As vehicles enter the realm of computers, more and more aspects of the vehicles will be automated every year, eliminating the need for drivers to control every function of the car.
This paradigm shift will require precise and accurate mapping solutions to keep the vehicle from driving off the road or running into the wrong lane. Soon, computers will steer cars across the world. DeepMap’s mapping technology is going to be at the center of these systems.”