Tsuyoshi Sato PIONEER CORPORATION July 6, 2017

Transcription

1 Technology R&D for for Highly Highly Automated Automated Driving Driving Tsuyoshi Sato PIONEER CORPORATION July 6, 2017

2 Agenda Introduction Overview Architecture R&D for Highly Automated Driving Hardware Architecture Cloud Architecture Key technology 3D-LiDAR Highly accurate self-localization of a vehicle Map data for Automated driving Summary

3 Overview/Architecture

4 Conventional/ADAS Vehicle ADAS+/Automated Vehicle(1 st Gen) Automated Vehicle(2 nd /3 rd Gen) Transition of Automotive technology Technology for human Provide the trust of automated driving Integrated Cockpit AVN for ADAS+/Automated 1 st Gen AV/Rendering Infotainment Vehicle Control AV NAVI Meter cluster vehicle Control Driver Monitoring HMI for ADAS+/Automated 1stGen Integrated Cockpit (partial) Object sensing Central control (ADAS) Driver monitoring HMI/BMI Context awareness Technology for machines Provide reliable automated driving Central control (Automated driving) Environment sensing Vehicle control Safety/Security/Privacy Vehicle control Confidential

5 Total System:1 st /2 nd Generation Automated Driving [ ] Object Recognition Front/Rear 3D-LiDAR Surround 3D-LiDAR 3D-LiDAR 3D-LiDAR 3D-LiDAR Camera Camera Camera Radar Ultra Sonic RAW AVB Driver Monitoring Camera Bio-Sensor Vehicle Status 3D-LiDAR Object Detection Module [HW/SW] Detection Module Detection Module Detection Module Driver Monitoring Module [HW/SW] Accel Braking Steering Object CAN-FD CAN-FD CAN/LIN Localization G/Gyro HMI GNSS Localization Module [HW/SW] Object Object Object Object Driver State Control State Central Control Sensor Fusion Path Planning Lv.1 Module [SW] Central Gateway Location Predic tion Path Planning AVH+N Advanced Map Cache Path Path Planning Lv.2 CAN-FD (Router & Security) Autonomous Driving ADAS Map Path Planning Lv.3 ADAS Function (Fail-safe) Gate Way Router & Security Probe Decision Selection /Fusion ASIL-D ICT DCM [LTE /4G] ITS [V2X] Control Planning ASIL-D /D Cloud/ Map Download Data for Localization Data for Path Plan Map Creation/ Update Update /Probe Data Advanced Map [Data] Vehicle Control Domain Gate Way CAN Traffic Info Vehicle Probe Brake ECU Engine ECU Transm. ECU EPS ECU Driver display Display HUD ASIL-D Confidential

6 Total System:3 rd Generation Automated Driving [ ] Integrated Cockpit Cloud/ Display Driver Monitoring Unit Camera Bio-Sensor Front Display Driver Monitoring Module [HW/SW] Rear Display HMI Unit AUI Feely Haptec AV Driver Display AV/Rendering Unit Multimedia MAP Driver Info Map/Localize for Driver Rendering HUD Meter/Cluster Status Map Download Data for Localization Data for Path Plan Map Creation/ Update Update /Probe Data Advanced Map [Data] Traffic Info Vehicle Probe Sensor Front/Rear 3D-LiDAR Ether-TSN Central Control Ether-TSN CAN-FD/LIN ICT DCM [5G] 3D-LiDAR 3D-LiDAR 3D-LiDAR Sourround 3D-LiDAR Camera Camera G/Gyro GNSS Vehicle Status Accel Braking Steering ASIL-D Control RAW RAW Sensor Status Localizer Localization Cognitive Central Gateway (Router & Security) Sensor Fusion Recognition Object Perception Sensor Fusion Prediction Long range Path Planning Lv.1 Mid range Path Planning Lv.2 Mission Plan Path Planning Lv.3 Advanced Map Cache Traffic Info Update /Probe Motion Plan Control Planning Stability Control ASIL-D ITS [V2X/V2V] Vehicle Control Brake ECU Engine ECU Transm. ECU EPS ECU ASIL-D Confidential

7 Overview of Cloud System Other Vehicle V2X Driver Monitoring Service Traffic Information Service Operating control Service Cloud System-A Cloud System-B Cloud System-C FOG/BaseST A B C Government /Authority 100m msec Gateway Administration Receiving Analytics Sending Data Storage Gateway Gateway Administration Receiving Analytics Sending Data Storage Gateway Gateway Administration Receiving Analytics Sending Data Storage Gateway Road/Infra Maintenance Maintenance Vehicle Ego Vehicle Local Area Sec/Min Regional Area Min/Hour Global Area Day/Month MAP company Sensor Probe Data Map Object Localization Pathplaning Driving Information Objects Driver info Obstacles Path Plan Location info Congestion Traffic Rule Topology Battery info Diagnoses Accident risk Fault prediction Failure analysis Insurance Company Maintenance Company

8 Key Technology

9 Example of Key Component/Function & Requirement Sensor Environment Perception Localization Path Planning Vehicle Control Cloud Map Data LiDAR Sensor Fusion Ego Motion Path/Planning [Mission] Control Planning Map Data Traffic Rules Camera Segmentation Dead Reckoning Path/Planning [Behavior] Vehicle Control Map Data Object Rader Tracking Scan Matching Path/Planning [Motion] Map Data Localization Ultra Sonic Classification Map Data Path plan FPGA/ASIC Low Heat ASIL_B/D Jamming/ Spoofing GPU/SoC/FPGA/ASSP Low Power (<15W) / Low Heat Safety :ASIL_B/C/D Security :Falsification, Hacking, Privacy Decentralized system / Centralized system Implementation :Deep Learning / Rule base GPU/ASSP Falsification Leakage

10 Key Technology Sensor 3D-LiDAR

11 Role of 3D-LiDAR LiDAR Three roles of LiDAR Advanced map Vehicle Curbstone Landmark Matching Object detection Linking with cameras, etc. Linking with GNSS/Gyro, etc. Linking with high-precision maps (1) Surrounding environment detection (2) High precision Vehicle location (3) Map creation and updating Lane marker No collisions Smooth driving Adjusting to latest conditions 11/28

12 Built-in Layout Image (Tentative) Built-in LiDAR in a headlamp unit (example) High-mount, inside cabin (example) Multi-sensor configuration for omni-directional sensing

13 LiDAR Technical Features Scanning of driving space MEMS* mirror and optics technologies enables 3D spatial scan of vehicle surroundings. Small angular divergence of laser beams realizes high angular resolution and it provides detailed 3D structure of the environment. (1) Scanner optics/mems (2) System TXRX/DSP/data processing target omni-directional optics MEMS mirror controller circulator beam shaper LD TX trigger pulser DSP & data processing pointcloud bandpass filter collector lens APD RX digitizer Direct time-of-flight (TOF) detection Cost-effective architecture (single LD/APD + DSP) Unique scanner technology Unique system technology target #1 target #2 1 st peak (8.2σ) 2 nd peak (3.9σ) Single laser ray scanned by MEMS With well-calibrated likelihood information (without thresholding)

14 Key Technology Localization High accurate self-localization

15 Necessity of self-localization Vehicle self-localization with map data is important for Automated Driving To perform appropriate path planning including lane change If current driving lane is uncertain, path planning may fail To comply with traffic law Need to stop at correct stop line if traffic signal indicates red To control vehicle appropriately at intersection, curve, etc... Vehicle position/pose is critical when driving an intersection/curve To upload high quality probe data If localization is inaccurate, the probe data quality will degrade even if expensive sensor is used. If accuracy of self-localization is high, reliability/quality of sensing data is increased and it facilitates detection of a discrepancy between real world and map data.

16 Localization Methodology Point cloud-based complex environment (e.g., city area) Landmark-based traffic signs, lane mark (e.g., highway) GNSS-based few structures (e.g., wilderness) Point Cloud-based Scan Registration localization, for example, Normal Distribution Transform (NDT) Iterative Closest Point (ICP) Pros: Easy to create data Cons: Data size is relatively big Landmark-based Extended Kalman Filter (EKF) based localization Pros: Relatively small data size Cons: Ambiguity problem if similar landmark is located near by

17 Key Technology Advanced MAP MAP Data for Automated Driving

18 Map contents required for automated driving for Navigation Road/Lane Network Traffic Rules for Navigation Road Geometry Traffic Rules for Vehicle Control Reference Diving Path for Vehicle Control Stationary Obstacles Localization Data

19 Key function processing using sensor and map data Map data Sensor Localization object data Localization GPS, LiDAR, Camera, Rader Stationary obstacle, Road geometry data Obstacle Detection LiDAR, Rader, Camera Moving object Detection LiDAR, Rader, Camera Traffic Rules, Reference driving path Path Planning

20 Summary

21 Summary Hypothesized the total system of the automated driving Looked over the whole functions for automated driving Key functions have following requirements to be implemented Low power, Low heat, Security, Safety, Privacy Deep learning and Rule base are both important for implementation of the key functions Realization of function and achievement of the requirements 3D-LiDAR and Map are key components for automated driving Highly accurate self-localization of vehicle is one of the key functions