Self-Healing Maps for Autonomous Driving Sanjay Sood GPU - PowerPoint PPT Presentation

Self-Healing Maps for Autonomous Driving Sanjay Sood GPU Conference| May 10, 2017

HERE T Technologies 7,000+ 7, Countries mapped Millions Mi ons of 200 20 changes made to the Employees in 56 countries map eve very day focused on delivering the world’s best map and location services Today, HERE provides services to nearly all global OEM brands as well as many market leading brands such Microsoft, Amazon, Facebook etc. 4 of of 5 30+ 30 In-car navigation systems in Europe and N America use HERE maps Years of experience transforming mapping technology

HERE T Technologies We are making sense of the world through the lens of location, helping people achieve better outcomes through data-driven location solutions

Maps for cars by cars Maps for cars

HERE’s HD Live Map solves several major problems Local knowledge of the rules of the road Enhanced s sensor f functi tionality ty for contextual awareness of the environment Planning of vehicle control maneuvers beyond s sensor v visibility ty Precise p positi tioning for lateral & longitudinal control

The industry has evolved, providing new possibilities for cars Connected Intelligent

Building the high definition map GPS Antenna LiDAR 700,000 Points Proprietary cameras 41.3º Per Second 96.4 MP every 6 meters Novatel GPS/INS

A high definition, continuously updated map requires the crowd High definition map Harness crowd- For near real- sourced sensor data time updates

Map Mach chine learning will be cr critica cal for self-driving ca cars 11

The self-healing map process & ingestion 2 the cloud 3 1 new feature 4 Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Drive paths markings…) Change detection Map Internal HD Pu Publish Aggregation Features Observations update map HD HD Live Map SENSO SE SORIS SDRI SD Sensor collection Aggregation in Creation of a Update map & (C (Campaign ma manageme ment) t) publish OEM Cloud OEM sensor observations HD map tiles

The self-healing map process Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Drive paths markings…) Change detection Map Internal HD Pu Publish Aggregation Features Observations update map HD HD Live Map SENSO SE SORIS SDRI SD (C (Campaign ma manageme ment) t) OEM Cloud OEM sensor observations HD map tiles

The self-healing map process Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Drive paths markings…) Change detection Map Internal HD Pu Publish Aggregation Features Observations update map HD HD Live Map SENSO SE SORIS SDRI SD (C (Campaign ma manageme ment) t) OEM Cloud OEM sensor observations HD map tiles

The self-healing map process Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Drive paths markings…) Change detection Map Internal HD Pu Publish Aggregation Features Observations update map HD HD Live Map SENSO SE SORIS SDRI SD (C (Campaign ma manageme ment) t) OEM Cloud OEM sensor observations HD map tiles

The self-healing map process Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Drive paths markings…) Change detection Map Internal HD Publish Pu Aggregation Features Observations update map Protobuf NDS HD Live Map HD SE SENSO SORIS SD SDRI (Campaign (C ma manageme ment) t) OEM Cloud OEM sensor observations HD map tiles

Cloud and edge analytics: What to do where, best? Polylines (lane markings, road edges…) Many to one Points (signs, signals, pavement Localization Localization Drive paths markings…) Change detection Change detection Map Internal HD Pu Publish Aggregation Features Observations update map Protobuf NDS HD HD Live Map SENSO SE SORIS SD SDRI (C (Campaign Campaign mngt. & fleet requests ma manageme ment) t) OEM Cloud HD map tiles OEM sensor observations

Summary 1 Vehicles today are connected and becoming more intelligent 2 Intelligent cars need a high definition map 3 An HD map needs the ability to heal itself via the crowd 4 Self-healing maps require large amounts of vehicle sensor data 21