elrob meeting09/01/061 elrob meeting 9.1.2006. elrob meeting09/01/062 agenda 11h00 introduction –...

ELROB Meeting09/01/06 1

ELROB Meeting 9.1.2006


Agenda

11h00 Introduction – (Roland)11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break -> Photo of Team

14h00 Parallel sessions with sub-teams - Navigation room ME A 31 – (Pierre) - Scene interpretation ME B 10 – (Cyrill) 15h00 Coffee Break 15h15 Sub-team presentations, ~10min each 15h40 Merging - (Roland) 16h15 Summary - (Roland, Sascha, Pierre)


Goals of the meeting

Precise what can be done given the available time and resources

Define the interfaces between the different modules

Define responsibility for each task


Disclaimer

We, members of the Smart Team, in recognition of the importanceof our technologies in affecting the quality of life throughout theworld, and in accepting a personal obligation to our profession andits members, do hereby commit ourselves to the highest ethicaland professional conduct and agree:

- to accept responsibility in making engineering decisions consistent with the safety, health and welfare of the public

- to disclose promptly factors that might endanger the public or

the environment - to reject any development in relation with any kind of

offensive actions

The Challenge

Date/Location: 15 - 18.5.2006, Hammelburg, Germany 15.5: Preparation and testing 16.5: Runs on separate tracks, presentation of capabilities beyond the defined scenarios. 17.5: Runs and trails in urban scenarios 18.5: Runs and trails in open terrain scenario Overlap with ICRA 2006 !

Goal: Overview of the European state-of-the-art in the field of UGVs (unmanned ground vehicles). Navigate in a predefined, realistic scenario in an intelligent manner ELROB is a field test of UGVs, not a competition

Scenario Urban and open terrain scenario The scenario details (route definition) will become available in advance over time

Preliminary general description of the scenario and the route Detailed description of the scenario and the route Map and personal inspection of the route through the team leader prior to the ELROB

Preparation on a test track Demonstration on actual test track can be repeated

The other teams Mainly specialists for all-terrain vehicle See www.elrob2006.org

The Open Terrain Scenario

Name: Surveillance and Reconnaissance in open terrain Terrain: Outdoor, mixed countryside Task:

drive pre-designated route (GPS via points?) detect points of interest (static and dynamic), identify, locate and record, report results to commander return to station in timescale

Environment: the maximum route is no longer than 2000 meters. includes paved roads, unpaved roads, trails, and off-road desert areas examples of obstacles include ditches, berms, washboard, sandy ground,

standing water, fire, rocks and boulders, narrow underpasses, construction equipment, concrete safety rails, power line towers, barbed wire fences and cattle guards.

In addition to the existing natural obstacles, the organizers might place obstacles (e.g. military equipment) on the route that may disable a vehicle if struck. These obstacles must be detected and circumnavigated by a vehicle to successfully complete the route. The terrain is wide enough for vehicles to bypass these obstacles if necessary.

The Team

Name: Smart-Team Vehicle: SmarTer (Smart All Terrain) Team members:

EPFL, Ecole Polytechnique Fédérale de Lausanne, Switzerland ALU-FR, Albert-Ludwigs-University of Freiburg, Germany

Sponsors / Supporters: RUAG Land Systems (Vehicle Design, Testing) Singleton Technology Sarl (3D Mapping) Com In Motion (Systems Design)

Team structure: Core-Team: > 50% of their working time Other Key Members

The Smart-Team

Wolfram Burgard

Roland Siegwart

Pierre Lamon

Sascha Kolski

Jan Weingarten Tarek Baaboura Marcelo Becker Viet Nguyen

Frederic Pont

Cyrill Stachniss

Björn Jensen

DaveFerguson

Luciano SpinelloGregoire Terrien

Beat Ott

Hanspeter Kaufmann

Shrihari Vasudevan

Kristijan MacekRudolph Triebel

Patrick Pfaff

Not on the imageDavide ScaramuzzaStefan GaechterGiorgio GrisettiOscar Martinez MozosChristian PlagemannAxel RottmannFerdi SessizGernot SpiegelbergArmin Sulzmann


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre)11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break



Project website

Public website: http://www.smart-team.ch

Private (internal use)http://www.smart-team/private/index_team.html login: “smart-team” password: “go_smart”

/home/plamon/Elrob/Html/index.html


Current state of the project

Navigation (cpu 0)Navigation (cpu 0)

3D Mapping (cpu 1)3D Mapping (cpu 1)

Scene interpretation (cpu 2)Scene interpretation (cpu 2)

Trajectory control Localization Path planning Obstacle negotiation Vehicle monitoring

Digital Terrain Modeling

Scene interpretation Objects recognition Segmentation

Car sensors and actuators

IMU + GPS

Navigation Sick

3D SicksStamped pose (6 dof's)

Simplified 3D map

OmnicamTraversability map (2.5 D)


Processing power

Description Compact PCI 19”, industrial (3 racks) Each rack contains at least

Pentium M @ 2GHz 1.5Gbyte of RAM 1 Gigabit and 1 Fast Ethernet 2 serial ports 30 Gbyte 2.5” hard disk 60 W, 24 V DC power supply


Processing power

Navigation rack (CPU 0) 4 channels serial interface board 2 independent CAN buses board

3D Mapping rack (CPU 1) 4 channels serial interface board 1 fire wire board

Scene interpretation rack (CPU 2) no sensor connected to this rack


Processing power

Not yet received But scheduled for january 16

Received this morning !

Still have to mount the racks in the trunk- mechanical hardware with anti-vibration system (Marcelo, Pierre, Sascha)

Install the OS- drivers etc. (Fred)


Hardware for the Smart

Mount the support on the roof to hold (Tarek)- the 3D sick- two navigation Sick- the omnicam

Mount the front sick Cabling 24V Power generator (RUAG) Lift the chassis (for rough terrain) (RUAG)


Steering

Principle

DA UnitDA Unit

Torque sensor

Vehicle velocity (Vehicle CAN)

Driver torque (voltage)

Madd

Default mode Steering angle sensor (CAN)


Steering

We use the driving assistance for steering

DA UnitDA Unit

Torque sensor

Velocity (CAN)

Torque(voltage)

Madd

ComputerComputer

Steering angle sensor (CAN)CAN to analogCAN to analog

Steering electronic board

Steering electronic board

Vehicle CAN

Com

pute

r C

AN

Veh

icle

CA

N

Switch

Position control (PID) Minimize :e = DesiredAngle – SteeringAngle


Steering

Enough torque to steer We could fake the unit Switch auto/manual works

- the system remains on

Test the PID (Sascha, Pierre) Redesign the board to be more reliable

- new pcb, clean cabling (Pierre)


Breaking system and gas

Hardware working Limited to full break/release

- proportional control (Sascha)

Switching between manual/auto working

video Use CAN to analog unit to control the gas


3D Sick

Description

It's turning! Cabling (Jan, Michael's brother) Add an encoder to get the angle (Jan, Tarek) Synchronization scan acquisition/rotation

(Jan, Fred)


IMU

Description roll, pitch and heading at 100Hz built in GPS (RTCM and WAAS) EMI and vibration resistant Environmentally sealed sensor fusion

GPS+accelerometers+gyros+compass

Usage use its outputs when full 3D GPS is available switch to wheel encoders + IMU when GPS is

not available (short period of time)


IMU

The unit has been received The interface has been coded It has to be properly mounted so that it is not

to much influenced by vibrations and perturbing magnetic fields

Extensive experiments to test its performance


Emergency stop

RUNRUN0

PAUSEPAUSE1

DISABLEDISABLE2

Stop Flash

T10

T01

T12

T02

{e_stop}

{e_stop}

{e_sw_run} {e_sw_pause}

Flash ON

Flash ON

Flash OFF

Wireless com

Remote control

[e_stop]

[e_stop, e_sw_run, e_sw_pause]


Emergency stop

Nothing available yet- RUAG might have a solution for wireless com ?

The state machine has to be implemented in a way that it is independent on an OS (reliability, reactivity, etc.)- micro-controller based unit with CAN bus and IO

(industrial standard). Takes full control of the car if the navigation RACK fails to regularly send signal. Directly receive wireless signal and E-stop.


Omnicam

Should stand on top of the 3D sick (aligned with the rotation axis)

Mechanical design (Tarek, Gregoire)option 1: tripod occlusion for the 3D sick

option 2: bearing with one leg for blocking the degree of freedom

Calibration (intrinsic/extrinsic) (Davide)


Software

Use of GenoM for the project

modular software architecture

communication via shared memory

reusable code

relatively easy to port on a specific architecture

Communication of large chunk of data between racks is too slow (long delay)!

- we have to use another mechanism for data transmission (IPC is a good candidate) (Fred)


First experiments

First experiments of autonomous navigation Credits: Sascha, Dave, Kristijan, ... Done with the “old” car (mechanical steering)

video


First experiments

Localization based on odometry and IMU


First experiments

Autonomous navigation based on D* and traversability maps


GenoM modules

= GenoM module

iCaniCan

Control- steering - gas- break

- vehicle speed - steering angle- etc.

SickSick

IMUIMU

vCanvCan

OdometryOdometry

Mappingglobal planninglocal planning

Mappingglobal planninglocal planning

Commandv, steering angle


Conclusion

There are still many open questions: Will the NAV420 be sufficient for localization ? When will be the 3D sick ready ? E-stop ? etc.

Many elements are already available We want a nice demo in the end In the first stage of the project, the focus

must be on integration !


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre)11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break



Milestones

first 3D scan + omnicam image available (static): end of january

SPARC demo (autonomous navigation in flat terrain): 4th of february

generator + chassis lift: middle of february 3D scans from the car: end of february E-stop ready: beginning of march integration week: 8-10 of march


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break



3D Mapping

What is available from the shelf ?

What can be developed within the remaining time period ?

What kind of info is needed for mapping and what can be provided to the Navigation and Scene interpretation modules ?


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams12h45 Definition of sub-teams

13h00 Lunch break



Scheme (first draft)




Trajectory control Localization Path planning Obstacle negotiation Vehicle monitoring

Digital Terrain Modeling

Scene interpretation Objects recognition Segmentation


IMU + GPS

Navigation Sick

3D SicksStamped pose (6 dof's)

Simplified 3D map

OmnicamTraversability map (2.5 D)


Afternoon discussion

Two teams Team A: Navigation (chair Pierre) Team B: Scene interpretation (chair Cyrill)

The goals are

1. draw a bloc diagram with inputs and outputs2. put responsible NAME for each bloc!3. prepare a presentation (powerpoint)



Team A: Navigation (chair Pierre)

Sascha Kolski Kristijan MacekFrederic Pont Tarek

BaabouraMarcelo Becker Hanspeter

KaufmannFredi Sessiz Gernot

SpiegelbergArmin Sulzmann Gregoire Terrien Jan Weingarten



Team B: Scene interpretation (chair Cyrill)

Patrick Pfaff Davide Scaramuzza

Luciano Spinello Cyrill Stachniss

Rudolph Triebel Stefan GaechterGiorgio Grisetti Bjoern JensenViet Tuan Nguyen Christian

Plagemann Axel Rottmann Shrihari Vasudevan Oscar Martinez Mozos


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break

14h00 Parallel sessions with sub-teams - Navigation room ME A 31 – (Pierre) - Scene interpretation ME B 10 – (Cyrill) 15h00 Coffee Break 15h15 Sub-team presentations, ~10min each15h15 Sub-team presentations, ~10min each 15h40 Merging - (Roland) 16h15 Summary - (Roland, Sascha, Pierre)


Agenda

11h00 Introduction – (Roland) 11h10 Current state of the project – (Pierre) 11h50 Project schedule and milestones – (Pierre) 12h00 3D Mapping discussion - all together 12h45 Definition of sub-teams

13h00 Lunch break

14h00 Parallel sessions with sub-teams - Navigation room ME A 31 – (Pierre) - Scene interpretation ME B 10 – (Cyrill) 15h00 Coffee Break 15h15 Sub-team presentations, ~10min each 15h40 Merging - (Roland)15h40 Merging - (Roland) 16h15 Summary - (Roland, Sascha, Pierre)


3D Mapping - Discussion

Discussion / Minutes

Simulator (Cyrill) would be a good thing to have Sensor observations

Needs 3D model of environment Could be nice for planning

Use: Verify the software on “sound” signals and timing issues

3D Mapping

Standard GPS -> Sufficient ? GPS alone is not sufficient ! Together with IMU it might be fine Precision of pose estimation:

Travel between two measures: around 2 m Global should be better than 5 m

Angular resolution of 3D measurement Sick-mirror rotation:

No interlace Reduced opening angle

In rotation of Sick platform: Scan alignment based on features (ALU-FR) Can we use omnicam for scan alignment? What is the minimal speed the Smart can move?

We have to check this Visual Motion Estimation probably as option Probably correction from 3D mapping to Navigation for “real” SLAM


Scene interpretation - Discussion

3D Mapping

Scene interpretation ? Should omni-cam be linked to cpu2? Output 3D map

2D Grid-map (10 cm) and multiple-Gaussians for surface heights (ALU-FR)

Should cpu2 get all new information.


Tasks for Scene Interpretation

Detecting distinctive objects [Luciano, Shrihari] Car/people from 3D map [Luciano, Rudi] Terrain classification (trees, bushes, roads, etc)

[Davide, Shrihari] Detecting moving objects [Cyrill (3D data),

Davide (vision)] Detecting buildings [Viet, Stefan,

Freiburg: Giorgio/Cyrill]


Required Input

Surface maps with wall information (2nd PC) Local point cloud with alignment to previous

scan (2nd PC) Data from the static SICK (1st PC) Pose information (1st PC) (Omni) camera should be connected to 3rd PC Do we need a perspective camera (connected

to 3rd PC) or a high resolution omni cam (ask a vision expert)?


Modules

Moving objectsMoving objects

Distinctive objectsDistinctive objects

People/car detectionPeople/car detection

Stamped pose (6 dof's) for all modules

Perspective camera ?

BuildingsBuildings3D Data (2nd PC)

Terrain classificationTerrain classification

3D Data (2nd PC)

3D Data (2nd PC)


Current state of the project





IMU + GPS

Navigation Sicks

3D SICKStamped pose (6 dof's)

Surface map with wall information;Aligned Point cloud

Traversability map (2.5 D) Stamped

pose and data of the static SICKs

Omnicam

Perspective camera ?


Navigation - Discussion


GenoM modules

= GenoM module

Low levelctrl..

(Kristijan,Sascha,Pierre)

Low levelctrl..

(Kristijan,Sascha,Pierre)

Control- steering - gas etc.

- vehicle speed - steering angle- etc.

NavSick(Fred,Jan

)

NavSick(Fred,Jan

)

IMU+GPS

(Pierre)

IMU+GPS

(Pierre)

vCAN(Sascha)

vCAN(Sascha)

Pose estimation(Sascha,Pierre)Pose estimation(Sascha,Pierre)

Motion Planning

(Sascha,Dave,Kristijan)

Motion Planning

(Sascha,Dave,Kristijan)

ECU(Sascha)

ECU(Sascha)

=Hardware module

E-stop receiver(Ruag ?)

E-stop receiver(Ruag ?)

Traversability map from 3D Mapping

DA(Pierre

)

DA(Pierre

)

elrob meeting09/01/061 elrob meeting 9.1.2006. elrob meeting09/01/062 agenda 11h00 introduction –...

Documents

maximum route

open terrain scenarioname

open terrain scenariooverlap

terrain vehiclesee

designated route gps

open terrainterrain

elrob preparation

realistic scenario