autoware: ros-based oss for urban self-driving mobility 2017 autoware.pdf · ros-based oss for...

44
Autoware ROS - based OSS for Urban Self - driving Mobility Shinpei Kato Associate Professor, The University of Tokyo Visiting Associate Professor, Nagoya University Founder and CTO, Tier IV Inc. https://github.com/CPFL/Autoware

Upload: vonhi

Post on 22-Mar-2018

305 views

Category:

Documents


20 download

TRANSCRIPT

AutowareROS-based OSS for Urban Self-driving Mobility

Shinpei KatoAssociate Professor, The University of Tokyo

Visiting Associate Professor, Nagoya University

Founder and CTO, Tier IV Inc.

https://github.com/CPFL/Autoware

Velodyne HDL-64e (3D LiDAR)

Point Grey Ladybug 5 (Camera)

IBEO LUX 8L (3D LiDAR)

Velodyne HDL-32e (3D LiDAR)

JAVAD RTK-GNSS (GNSS/GPS)

Point Grey Grasshopper3 (Camera)

Velodyne VLP-16 (3D LiDAR)

TorqueSensor

Gas PedalPositionSensor

ShiftPositionSensor

BrakeStrokeSensor

OilPressureSensor

Emergency Button

EPSECU Hybrid ECU

Skid ControlECU

OtherECUs

Control Box Status Monitor

CAN0CAN1

CAN I/F

OBDIIConnector

Direct signals to ECUs

bypassing CAN.

Computing Unit

Reconnected wires.

Car interior monitors.

Accessories.

CAN

Cameras3D LiDARs GNSS/GPS

NVIDIA DRIVE PX2

Aisan Technology’s Minivan (ZMP RoboCar)

Public road demonstration

Yamaha Motor’s Golf Cart

Yamaha Motor’s Baggy

31 miles along the El Camino Real in

the Bay Area along 140 traffic signals

and crosswalks during regular traffic

over a period of an hour-and-a-half.

Sensing Computing

Detection

Localization

Prediction

Perception

Decision

Planning

Motion

Mission

Intelligence State

LiDARs

Cameras

IMU

GPS

Actuation

YMC

ZMP

TMC

AS

Data Socket System Util

points_localizer

{lidar0,1,2,…}/points_raw

points_concat

dead_rekoner gnss_localizer

gnss_pose

map_manager

points_map_region points_map_updatedruntime_manager

static_pose

nmea_sentence

delta_poselocalizer_velocity

can_velocity imu_raw

current_pose

vel_pose_

connector current_velocity

Packages Example (Localization)

points_filtered

tf

manual_posetf

Aisan Technology (Partner Company)

High-Definition 3D Mapping

High-Definition 3D Mapping

3D map data Sensor scan data

Matching

Localization

Normal Transform Distributions (NDT) for Localization

Normal Transform Distributions (NDT) for Mapping

Euclidean Clustering for Detection

Convolutional Neural Networks (CNN) for Detection

Convolutional Neural Networks (CNN) for Detection

SSD Yolo2

Convolutional Neural Networks (CNN) for Detection

Depth Image

Height Image

Intensity Image

CNN Segments

Camera-LiDAR Calibration and Sensor Fusion

KITTI Dataset Plugin

Traffic Signal

Ego Vehicle

Traffic Light Recognition using 3D Map Information

A* Search for Path Planning (Trajectory Generation)

State Lattice for Path Planning (Trajectory Generation)

Pure Pursuit for Path (Waypoints) Following

Task Trace Tool (Ftrace)

DNN Prediction

Driving School Teacher

Driving School Teacher (details)b

Automanhttps://www.automan.ai

Brake

or

Not to Brake?

NVIDIA – AutonomouStuff – Tier IV

Sensing Computing

Detection

Localization

Prediction

Perception

Decision

Planning

Motion

Mission

Intelligence State

LiDARs

Cameras

IMU

GPS

Actuation

YMC

ZMP

TMC

AS

Data Socket System Util

NVIDIA Drive PX2

NVIDIA – AutonomouStuff – Tier IV

https://github.com/CPFL/Autoware

ROSBAG STOREhttps://rosbag.tier4.jp