Webinar1 darpa07

Self-Driving Car Webinar#1 ARCHITECTURE BREAKDOWN OF RACING VEHICLES IN

THE DARPA 2007 URBAN CHALLENGE

Mengchen Ma

Postgraduate student at EBS University, Germany, Class of 2015

M.Sc. in Automotive Management/ B.Sc. in Automotive Engineering

Previous: Roland Berger (Shanghai), Opel (Germany)

Focus Area: Self-Driving Car, Car Sharing

OverviewA. The DARPA 2007 Urban ChallengeB. Talos (MIT)C. Boss (CMU)D. Junior (Stanford U.)E. Summary

A. The DARPA 2007 Urban Challenge

Before 2007 Urban ChallengeDGC 2004 DGC 2005

No team travelled more than 7.4 miles Five teams completed the race

Finished

Final

NQE

Site Visit

Registered

6

11

35

53

89

2007 Urban Challenge Result1st PlaceBoss

2nd PlaceJunior

4th PlaceTalos

B. Talos (MIT)

Overall Architecture

Navigator

Drivability MapMotion Planner

Controller

Fast Vehicles

Hazards

Obstacles

Road Paint DetectionLanes

Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

Perc

eptio

n

CamerasCameras

RadarsRadars

SICKSICK

Sens

ors

Planning &

Control


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK

Sens

ors

Planning &

Control


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICKPlanning

& Control


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK

Obstacle Detection and TrackingSICK VelodyneSICKSICK SICKSICKRadarsSICKRIEGL

Data Association

Velodyne Front-endPlanar LiDAR Front-end

Spatial Clustering

Temporal Association

Least-square Trajectory Estimation

Stationary objects Velocity tracks

planar point clouds3D point

clouds

Off-ground point detections

Constant velocity model


Data Association


Spatial Clustering





clouds




Data Association


Spatial Clustering





clouds




Data Association


Spatial Clustering





clouds




Data Association


Spatial Clustering





clouds




Navigator


Controller

Fast Vehicles

Hazards

Obstacles

Road Paint Detection

Lanes

Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK

Lane Detection and Tracking

SICK

Velodyne

SICKSICK

SICKSICKCamerasRoad Paint Detection

Obstacles

Curbs

Lane Centerlines Estimator

Lane TrackingRNDF

Lanes


SICK

Velodyne

SICKSICK


Obstacles

Curbs


Lane TrackRNDF

Lanes


SICK

Velodyne

SICKSICK


Obstacles

Curbs


Lane TrackingRNDF

Lanes


SICK

Velodyne

SICKSICK


Obstacles

Curbs


Lane TrackingRNDF

Lanes


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Lanes

Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK

The Navigator is responsible for planning the high-level behavior of the vehicle including:• Shortest route to the next MDF checkpoint• Intersection precedence, crossing, and merging• Passing• Blockage replanning• Generation of the goal for the Motion Planner• Generation of the failsafe timers• Turn signaling


Navigator

Drivability Map

Motion Planner

Controller

Fast Vehicles

Hazards

Obstacles


Lanes

Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK


Navigator


Controller

Fast Vehicles

Hazards

Obstacles


Lanes

Vehicle Positioning

Radars

SICK

Velodyne

Cameras

IMU, GPS, Odometry

MDF RNDF

CamerasCameras

RadarsRadars

SICKSICK

RRT-Based Motion Planner

Input to the controller

Predicted trajectory

Stopping nodes

Obstacle

Road Departure → Infeasible

Divider Crossing→ Infeasible

Obstacle Collision→ Infeasible

Feasible Paths Goal

Car

C. Boss (CMU)

Overall Software ArchitectureVe

hicl

es

Perc

eptio

n

Mission Planning

Behavioral Executive

Motion Planning


hicl

es

Perc

eptio

n

Mission Planning


Motion Planning

Perception —Sensor SetupSensor Type Amount Max.

RangeVertical Angle

Horizontal Angle Returned Feature

Continental ARS 300

Scanning Radar(near/far

)5 60/200m 4.3° 56°/18° 2D position, 2D velocity

Continental ISF172

Fixed Beam Laser 2 150m 4° 14° 2D position

SICK LMS291

Scanning Laser, 1 level 6 80m 0.25° 180° Edge Target, 2D position

IBEO Alasca XT

Scanning Laser, 4 level 2 200m 3.2° 240° Edge Target, 2D position

Velodyne HDL-64E

Scanning Laser, 64 beam 1 120m 26.8° 360° Edge Target, 2D position,

Obstacle Maps

Perception —Sensor Setup

Perception —Architecture

Dynamic Obstacle Removal

Instantaneous Obstacle

Map Generation

Object Prediction

Hysteretic Obstacle Map

GenerationObject

TrackingRoad Shape Estimation

Map Database

Static Obstacle MapDynamic Obstacle Hypothesis List

Road Structure

Static Obstacle Estimation Tracking Road Estimation

Perception —Architecture

Dynamic Obstacle Removal

Instantaneous Obstacle

Map Generation

Object Prediction

Hysteretic Obstacle Map

GenerationObject

TrackingRoad Shape Estimation

Map Database

Static Obstacle MapDynamic Obstacle Hypothesis List

Road Structure

Static Obstacle Estimation Tracking Road Estimation

Tracking Subsystem —Architecture

Sensor LayerLocal Classification & Proposal Generation

AssociationLocal Feature Validation

Feature Extraction

Fusion LayerObject/List Management Global Feature ValidationEstimation & Prediction Check Against

Road StructureModel SelectionGlobal Classification

Dynamic Obstacle Hypothesis ListRoad Structure &Instantaneous Map

FeaturesValidatedFeatures

Measurement(Observations, Proposals,Movement Observations)



Feature Extraction



Feature Extraction


hicl

es

Perc

eptio

n

Mission Planning


Motion Planning

Behavioral Executive—Architecture

Transition Manager

Precedence Estimator

Vehicle Driver

Current Scene Reporter

State Estimator

Goal Selector

Lane Driving

Lane Selector

Distance Keeper

Merge Planner

Goal Selection

Intersection Handling

Data in from System

Data out to System

Behavioral Executive—Goal Selection


Vehicle Driver


State Estimator

Goal Selector

Lane Driving

Lane Selector

Distance Keeper

Merge Planner

Goal Selection

Data in from System

Data out to System

Transition Manager


Behavioral Executive—Intersection Handling


Vehicle Driver


State Estimator

Goal Selector

Lane Driving

Lane Selector

Distance Keeper

Merge Planner

Goal Selection

Data in from System

Data out to System

Lpolygon = Vmax(Taction+Tdelay+Tspacing)+Dsafety

Lpolygon = Vmax (Taction+ Tdelay+ Tspacing) + Dsafety

Transition Manager


C. Junior (Stanford)

Overall Software Architecture

Overall Software ArchitectureSe

nsor

s

Sensor Overview

Bumper Lasers (2*IBEO, 2*LDLRS)• Vehicle Tracking

3D Laser (Velodyne)• Vehicle Tracking• Obstacle Detection

2*SICK & 2*RIEGL Lasers• Lane & Curb Localization

5*Radars (Bosch LRR2)• Merge Safety

Overall Software Architecture

Mapping&

Localization

Basic Concept of SLAM SLAM: Simultaneous Localization And Mapping

Mapping: Construct a map of unknown environment

Localization: Keep track of robots’ location relative to the map

Applications: Indoor/Outdoor Robots, Underwater/Airborne systems

Typical methods: Kalman Filter / Particle Filter

Basic Process of SLAM

EKFOdometry

update

EKFRe-observation

EKFNew

observation

Data Association

Landmark Extraction

Laser Scan

Odometry Change

EKFOdometry

update

EKFRe-observation

EKFNew

observation

Data Association

Landmark Extraction

Laser Scan

Odometry Change

Robot

Landmark

EKFOdometry

update

EKFRe-observation

EKFNew

observation

Data Association

Landmark Extraction

Laser Scan

Odometry Change

EKFOdometry

update

EKFRe-observation

EKFNew

observation

Data Association

Landmark Extraction

Laser Scan

Odometry Change

EKFOdometry

update

EKFRe-observation

EKFNew

observation

Data Association

Landmark Extraction

Laser Scan

Odometry Change

RNDF Localization

Lateral Posterior Distribution

Lane Line Distribution

Applanix Localization

E. Summary

SummarySIMILARITIES

• Sensing: Multiple sensors, partially fused, partially redundant, and also with specialized functions. (First appearance of 3D LiDAR in autonomous vehicles)

• Perception: Sensor/Data fusion, mapping and localization, obstacle detection and tracking

• Planning: Mission planning, motion planning, (path planning)

NOVEL FEATURES

• Boss: Various components from behavioral subsystem (Lane driving, goal selection, intersection handling)

• Junior: Precise localization with the help of road reflectivity and curb-like obstacles

• Talos: A unified planning and control architecture (Navigator, motion planner and drivability map)

References Darms, M. S., Rybski, P. E., Baker, C., & Urmson, C. (2009). Obstacle detection and tracking for the urban challenge. IEEE Transactions on Intelligent Transportation Systems, 10(3), 475-485.

Leonard, J., How, J., Teller, S., Berger, M., Campbell, S., Fiore, G., ... & Koch, O. (2008). A perception driven autonomous urban ‐vehicle. Journal of Field Robotics, 25(10), 727-774.

Levinson, J., Askeland, J., Becker, J., Dolson, J., Held, D., Kammel, S., ... & Sokolsky, M. (2011, June). Towards fully autonomous driving: Systems and algorithms. In Intelligent Vehicles Symposium (IV), 2011 IEEE (pp. 163-168). IEEE.

Levinson, J., Montemerlo, M., & Thrun, S. (2007, June). Map-Based Precision Vehicle Localization in Urban Environments. In Robotics: Science and Systems (Vol. 4, p. 1).

Montemerlo, M., Becker, J., Bhat, S., Dahlkamp, H., Dolgov, D., Ettinger, S., ... & Johnston, D. (2008). Junior: The stanford entry in the urban challenge.Journal of field Robotics, 25(9), 569-597.

Riisgaard, S., & Blas, M. R. (2003). SLAM for Dummies. A Tutorial Approach to Simultaneous Localization and Mapping, 22(1-127), 126.

Urmson, C., Anhalt, J., Bagnell, D., Baker, C., Bittner, R., Clark, M. N., ... & Gittleman, M. (2008). Autonomous driving in urban environments: Boss and the urban challenge. Journal of Field Robotics, 25(8), 425-466.

Thank you!#STAND ON THE SHOULDERS OF GIANTS

[email protected]

Antoni Ma

KiddGilchrist

Mengchen Ma