Seeker kick-off workshop

“State of the art”

Simon LacroixLaboratoire d’Analyse et

d’Architecture des SystèmesCNRS, Toulouse

Disclaimer

• Learn last Thursday that I was to give this talk…• Was in Barcelona Thursday / Friday…

• Had a week-end already “burnt” by work

• Plus:• Have not worked on planetary rovers since 2005• Not aware of latest Exomars developments• Knows nothing on Mars Science Laboratory navigation (but who

else than JPL knows?)

What state of what art ?

• MER missions• 2004 baseline stems from mid 90’s research• Has been used as a testbed for improvements

(visOdom, global Nav with D*, target reaching)

• Exomars / MSL navigation approaches

• Publications related to planetary rover navigation

• Huge progresses / publications in vision & navigation in the robotics community

Seeker main axes of development

• Long range navigation

• Mission planning

• Science autonomy

A glimpse at recent publications

• Other essential information sources: • ICRA and IROS space robotics workshops• ASTRA and I-SAIRAS conferences

Related to seeker

• Journal of Field Robotics• 2007: 3 special issues on “Space Robotics”• 2009: 2 special issues on “Space Robotics”• More to come end of this year

• 21 papers• 7 not on rovers (out of Seeker scope)• 2 on rover chassis conception• 4 on locomotion (advanced motion control)• 2 on localization• 2 on navigation• 3 on high level planning• 2 on “target reaching”

Seeker main axes of development

• Long range navigation

• Mission planning

• Science autonomy

Navigation

Three main functionalities:

Perception: data acquisition, environment map building, localization

Decision: path and trajectory determination

Action: locomotion control and monitoring

(plus overall control / supervision)

MER approach

· Main characteristic: validated !· Overview: simple loop

1. Data acquisition: stereovision

2. Environment modelling: local navigation map

3. Decision: elementary trajectory evaluation

4. Trajectory execution (short distance)

· Localization: odometry + IMU (+ sun sensor)· Overall control: sequential loop

MER approach

stereovision

motion selection

Locomotion,then stop

traversability analysis

MER “extended” approach

· VisOdom

· “Itinerary planning” using D*

· Target reaching

CNES approach

· Main characteristic: experimentally validated· Overview: a bit more complex loop

1. Data acquisition: stereovision

2. Environment modelling: navigation map updating

3. Decision:

1. Sub-goal determination

2. Trajectory planning

3. Perception planning

4. Trajectory execution (limited distance)

· Localization: odometry + IMU· Overall control: sequential loop

CNES approach

stereovisiontraversability analysis

Locomotion,then stop

robot

Sub-goal, trajectory and perception task selection

MER vs CNES approaches

In 2004:

· MER approach– The simplest, local navigation (“Obstacle avoidance”

scheme)

· CNES approach– More global reasoning (“Navigation planning”)

Þ Able to more efficiently deal with longer missions

Þ This comparison does not hold since D* integration on MER

(LAAS approach)

Overall loop: sub-goal, trajectory, perception task and motion mode selection

Easy terrain mode Rough terrain mode Mode n

…

But: certainly too complex for Seeker

Lessons learned from MER

· Many easy situations tackled under direct control

· Localisation is a critical issue in some situations

· Better locomotion abilities required (slip detection)

Rover Autonav distance Total distance

Spirit 1253 m 3405 m

Opportunity 224 m 1264 m

As of June 14, 2004 :

Lessons learned from MER

· Higher autonomy required to traverse cluttered areas

· Several ground interactions required to place the instruments

· End of autonomous motions may leave rover in bad attitude (wrt. antenna, solar panel)

Localization

· 3D odometry integrates:– Wheel encoders

– Wheel steering angles

– Chassis angular configuration

· Inertial localization: assess among the following possibilities– Attitude information (at stop and during motions)

– Heading provided by the integration of a gyrometer (what drift? Are there space qualified FOGs ?)

– Integration of 6-axis IMU - fused with odometry

Vision-based localization

· Visual odometry principle

Stereo : 3D points Stereo : 3D points

Motion estimationwith matched 3D

points

Tk Tk+1

Vision-based localization· Numerous progresses since 2004:

· Feature-based SLAM· Single cams, stereo cams, panoramic cams· Efficient EKF or optimization solutions· INS / odometry integration

Vision-based localization· Numerous progresses since 2004:

· Appearance-based localization· E.g. fabMap @ Oxford· Efficient way to detect loop closures

Locomotion control and monitoring· Opportunity in April 2005

Locomotion control and monitoring· Opportunity in April 2005

Locomotion control and monitoring

Locomotion monitoring of essential importance– Position tracking– Attitude and chassis internal angles checking

· Wrt fixed thresholds· Wrt to a “configuration space trajectory”

– Wheel slippage detection– Localisation algorithms monitoring

Plus: Recovery procedure definition required

Essential importance of diagnosis (FDIR)(non only locomotion, but overall navigation)

Seeker main axes of development

• Long range navigation

• Mission planning

• Science autonomy

ESA-driven GOAC project

“Goal Oriented Autonomous Controller”:

Goal-oriented operations:A goal tells the robot what to do, instead of how to do it.

Model-level programming:High-level of abstraction raising the focus to the problem domain: features of the robot mechanisms, available behaviours, task domain.

Robust execution in an uncertain environment:Let the robot decide how best to accomplish an objective, based on situation at hand.

Safe execution:The model used by the planner can capture safety constraints, so that all plans produced are guaranteed to comply with these constraints.

Correct-by-construction functional layer:Only allowed behaviours will be actually executed.

ESA-driven GOAC project

DELIBERATIVE LEVEL

º

TM ACQUISITION AND PROCESSING

SCIENCE DATA

ASSESSMENT AND

PLANNING

SYSTEM DATA ASSESSMENT

PLANNING

ROVER ACTIVITYPLANNING, VALIDATION

AND COMMAND GENERATION

TM

IMAGESIMAGES

NAVIGATIONPRODUCTS

HK PRODUCTS

PLAN EXECUTION REPORT

PRELIM. SCI . PLAN

ACTIVITY PLAN/COMMAND PRODUCTS

SIMULATION DATA

COMMAND PRODUCTS

ROVER STATUS

EVENTS AND COMM.

SKELETON

ON-BOARD SW MANAGEMENT

MEMORY IMAGES

POST-MISSION PRODUCT

GENERATION

SIMULATOR

EQM/MTS

PRELIM. ENG. PLAN

ROVER OPERATIONS CONTROL SYSTEM

EXOMARS MISSION

OPERATIONS CENTRE

FUNCTIONAL LEVEL

EXECUTION CONTROL

SENSORS AND ACTUATORS

NAVIGATION COMMUNICATIONS

POWER

FDIR

PASTEUR P/L

P/L SUPPORT EQUIPMENT

ROVER

GROUND STATIONS

THERMAL

TVCRPLANNING

H/W INTERFACE

MARS ORBITER(MRO)

TM/ TC

TM/ TC

TM/ TC

The level of uncertainty is extremely high

Planning involves ground assessment. Autonomous path

planning is a must!

Communications are scarce due to a narrow comms window and long

distance (latency)

Plan has to merge science activities with engineering

activities

Plan has to be validated by using the Simulator, before being

uplinked

The tactical operation’s process has very strict deadlines that have to be accomplished on a per-sol basis

As a conclusion

Ongoing R&D issues (according to L. Matthies, 2005)– Much faster flight processor

– 3-D perception at night

– Parachute for high elevation landing

– Brushless motors

– Nuclear power

– Locomotion mechanisms for very steep terrain

– Landmark recognition during descent

– Single command target approach and instrument placement

– Path planning for rough and steep terrain

– Position estimation on slippery terrain

– More automated long range position estimation and mapping

– More automated sequence generation for mission planning

Not related to Seeker

Very relevant – but not in Seeker’s scope

Definitely targeted by Seeker, many solutions exist in the literature