36
1 Localization and perception for autonomous navigation using automotive sensors Philippe Bonnifait Professor at the Université de Technologie de Compiègne, France Heudiasyc UMR 7253 CNRS Autonomous Driving Technology Radisson Schwarzer Bock Wiesbaden, Germany 21st - 22nd May 2014

Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

1 Ph. Bonnifait

Localization and perception for autonomous navigation using automotive sensors

Philippe Bonnifait

Professor at the Université de Technologie de Compiègne, France

Heudiasyc UMR 7253 CNRS

Autonomous Driving Technology Radisson Schwarzer Bock Wiesbaden, Germany

21st - 22nd May 2014

Page 2: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

2 Ph. Bonnifait

Outline

1. Renault valet parking robot

2. Autonomous electrical vehicles

3. Navigation, perception and localization systems

4. Integrity issues

5. Conclusion

Page 3: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

3 Ph. Bonnifait

Purpose An integrated valet service for the use

of EVs within the Technocentre Renault with autonomous driving technologies.

Objectives To provide a System Solution

which includes an ecosystem for the use of computer controlled EVs evolving in constrained spaces

To develop safe and reliable systems for autonomous vehicles, using automotive type components

To build technological modules that constitute the basis for intelligent mobility

Maintenance

Battery Charging

Automated

Valet Parking

Manual Autonomous

(Standby)

Power-off

Operational Mode

Automatic Switching

Manual Switching

The Renault valet parking project

Page 4: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

4 Ph. Bonnifait

Vehicle parking area

Pick-up points

An integrated Autonomous Driving Service

Page 5: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

5 Ph. Bonnifait

The valet parking robot

Page 6: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

6 Ph. Bonnifait

Scientific and technological goals of the project

- Study and experiment a valet functionality for an electrical automotive vehicle using automotive type components

- Develop prototypes based on standard electrical vehicles

- Identify good information processing methods in terms of localization, perception, navigation, control, decision and integrity monitoring

- Gain experience on the expected performance given a sensor suite configuration

Page 7: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

7 Ph. Bonnifait

Autonomous vehicles

Page 8: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

8 Ph. Bonnifait

Autonomous system overview

Communications Module

Integrity Supervisor

Map

Page 9: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

10 Ph. Bonnifait

Navigation

- Generates trajectories for Reaching the mission goal Avoiding obstacles and stay on the drivable space

Page 10: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

11 Ph. Bonnifait

Architecture

Global Planning System

Local Planning System

Navigation System

Supervisor control

Goal

Perception system

Localisation System

Global Trajectory

Local Trajectory

Control system

Map Vehicle Model

Page 11: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

12 Ph. Bonnifait

Perception sensors field of view

Page 12: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

13 Ph. Bonnifait

Perception System Output

Body frame

Coherent detections are fused together

Page 13: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

14 Ph. Bonnifait

Localization System

-Components CAN bus proprioceptive data GPS receiver Mobileye camera Map of the drivable area

-Output (10Hz) Position and heading Speed Confidence indicators

Page 14: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

15 Ph. Bonnifait

Localization system « key ideas »

- No IMU Very different from existing systems The key idea is to rely on proprioceptive sensors already installed on board the vehicle

- It is automotive-integrated (no Velodyne used and no additional encoder installed on a wheel)

- Try to be as much as possible GPS standalone no additional GNSS service (e.g. DGPS, omnsitar)

Page 15: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

16 Ph. Bonnifait

Localization system « key ideas »

- It exploits sensors used for other functionalities e.g. - The Mobileye camera initially installed for obstacle detection/recognition, - The ESP yaw rate sensor.

- Enhanced map with geo-referenced lane markings

-Thanks to the use of CAN-bus dead-reckoning, the system is able to handle short GPS outages and short sections without lane marking for instance at cross-roads. The availability for the navigation system is high.

Page 16: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

17 Ph. Bonnifait

Frames

Page 17: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

18 Ph. Bonnifait

GPS measurements

X

λ

Y

Z

φ

z

x

y

O

Page 18: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

19 Ph. Bonnifait

Localization solver

Page 19: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

20 Ph. Bonnifait

GPS measurements

Page 20: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

21 Ph. Bonnifait

GPS errors

GPS positioning errors are not white and time-correlated • Pseudo-range measurements are affected by atmosphere biases, • Satellites positions used in real-time ephemeris are inaccurate, • Position fixes are computed by a Kalman filter that has its own dynamic.

Page 21: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

22 Ph. Bonnifait

GPS errors modeling

- first order auto-regressive process

- random constant

- Shaping filter

Page 22: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

23 Ph. Bonnifait

Evolution model

Page 23: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

24 Ph. Bonnifait

Complete GPS observation model

Page 24: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

25 Ph. Bonnifait

Camera measures

Parameter name Definition

LaneQuality Quality

C0 Position Parameter

C1

Heading Angle

Page 25: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

26 Ph. Bonnifait

Observation model

Page 26: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

27 Ph. Bonnifait

Map-Matching camera measurements

- Select the correct segment [AB] from the map

- Two main stages Lane marking type selection Metric and side selection

Page 27: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

28 Ph. Bonnifait

Map-matching properties

Page 28: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

29 Ph. Bonnifait

Page 29: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

30 Ph. Bonnifait

ULS implementation - Sensor interfaces

NMEA GPS and Time Pulse (for synchronisation)

Vehicle can bus

- wheel speeds, yaw rate, steering wheel angle

- Mobileye lane markings

-Map OpenStreetMap and Spatialite format

- Algorithms GPS and CAN bus: C++

Map Management : C++

Map Matching: C++

Kalman filter real-time implementation: C++

Page 30: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

31 Ph. Bonnifait

Experimental results

Page 31: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

32 Ph. Bonnifait

Integrity issues

Page 32: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

33 Ph. Bonnifait

Integrity monitoring and fail-safe operation

- When navigating autonomously, the valet robot has nobody aboard

- A supervision system is needed to deal with hazardous situations

- fail safe state = vehicle stopped

- Integrity monitoring • Infrastructure supervisor • On-board the vehicle

Page 33: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

34 Ph. Bonnifait

Supervisor Box

Renault Vehicle Interface

Command and Mission Messages

HeartbeatMessages

V2V Comm Link

Operator Interface

Infrastructure supervisor

To ensure the external supervision of the autonomous vehicles and their interaction within the application environment

Page 34: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

35 Ph. Bonnifait

Autonomous integrity monitoring

- Perception system Object/obstacle missed-detections are crucial Method: - Use different sensors technologies and redundant fields of view - Check the state of the tracking perception systems - Every sensor report is kept (even if it is a false alarm)

- Localization system Reliable position confidence domains are used to check that the estimated accuracy is compatible with the required accuracy for the current navigation task Method: - Exploit data redundancy - Apply internal fault detection tests (GPS multi-path) - Robustify the map-matching algorithm

Page 35: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

36 Ph. Bonnifait

Conclusion

- Use of automotive type sensors for autonomous navigation - Robotic approach: the vehicle is controlled trough a localization feedback given a mission goal and can evolve in dynamic environments - An enhanced map and accurate sensors modeling is crucial - Many tests have been conducted by Renault on dedicated test tracks and different conditions - The performance of the perception/localization system is very encouraging by considering the automotive sensors that have been used More tests and validation are needed (accuracy and integrity)

- A “system engineering approach” is crucial for the design of such kind of system

Page 36: Localization and perception for autonomous navigation ...bonnif/talks/ADT_slides_Ph_Bonnifait.pdf · - Use of automotive type sensors for autonomous navigation - Robotic approach:

37 Ph. Bonnifait

Associated publications

-Tao, Z. and Bonnifait, Ph. and Fremont, V. and Ibanez-Guzman, J. « Mapping and localization using GPS, lane markings and proprioceptive sensors » IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Tokyo, Japon, pp. 406-412, Nov., 2013

-Tao, Z. and Bonnifait, Ph. and Fremont, V. and Ibanez-Guzman, J. « Lane marking aided vehicle localization » 16th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2013), La Hague, Pays-Bas, pp. 1509-1515, Oct., 2013

-Fouque, C. and Bonnifait, Ph. « Matching Raw GPS Measurements on a Navigable Map Without Computing a Global Position » IEEE Transactions on Intelligent Transportation Systems, vol. 13, num. 2, pp. 887-898, June, 2012