Rover Autonomy for Long Range Navigation and Science DataAcquisition on Planetary SurfacesTerry Huntsberger, Hrand Aghazarian, Yang Cheng, Eric T. Baumgartner,Edward Tunstel, Chris Leger, Ashitey Trebi-Ollennu, and Paul S. SchenkerJet Propulsion Laboratory, California Institute of Technology4800 Oak Grove Drive, Pasadena, CA [email protected]://prl.jpl.nasa.govIEEE International Conference on Robotics & AutomationCrystal Gateway Marriott Hotel, Washington, DCMay 15, 2002Rover Autonomy2T. Huntsberger et al., 05/02Contents• Introduction• FIDO Background• Long Range Navigation• Long Range Rendezvous• Single Command Science Target Rendezvous• Experimental Studies• Conclusions & Current DirectionsRover Autonomy3T. Huntsberger et al., 05/02Introduction• Reference mission for Mars SmartLander (MSL) 2009 calls for long,continuous, autonomous traverseson the order of 450 meters.• Numerous science sites separatedby as much as 3 kilometers areplanned.• Mission could last as long as 1000sols.• Greater onboard rover autonomy isneeded in order to maximize sciencedata return.Rover Autonomy4T. Huntsberger et al., 05/02FIDO (Field Integrated Design & Operations) Development Environment• Multiple JPL technology robots employ this architecture: the MER/Athena-inspired FIDO rover, MER Egress Rover, SRR, Inflatable Rover, RobotWork Crew, LEMUR, ATE, Cliff-bot...• The FIDO development environment provides reusable software (motioncontrol, stereo processing, guidance, manipulation, user-instrumentationinterfaces, etc.)• Operations of resulting rovers are based in a common mission operationstoolset: WITS (Web Interface for TeleScience), a distributed andcollaborative environment for planning-sequencing and data productdownlink, and Viz (ARC)• FIDO field experience to date has shown that these terrestrial system analogsreduce mission risk, providing cost-efficient integrated technologydevelopment, testing & evaluation within a flight-relevant environment, withdirect flight participationRover Autonomy5T. Huntsberger et al., 05/02Rover Autonomy6T. Huntsberger et al., 05/02• Mars Exploration Rover (MER)–mission simulations & science training inrealistic terrestrial environments for ops &scenario validation– WITS/Web Interface for TeleScienceselected as the MER science activityplanning tool–testing interfaces with MIPL for field trialtelemetry processing–targeted engineering and functional tests(instrument arm, localization repeatability)– MarsYard, Arroyo, & field tests in directsupport of the MER project– FIDO product transfers including personnel• Mars Smart Lander (MSL) & MarsSample Return (MSR)–advancement of “go-to” capability–enablement of visual rendezvous/return–development of mobile in situ sampling–technology benchmarking & reportingTechnology & Mission RelevanceRover Autonomy7T. Huntsberger et al., 05/02Example: FIDO Field TestGround Data System (GDS) InterfacesRover Autonomy8T. Huntsberger et al., 05/02PanCam ImagingNear-IRPoint SpectraFIDO Field Test Data ProductsNavCam PanoramaRock Micro ImageryBefore RATAfter RATSoil Micro ImageryRover Autonomy9T. Huntsberger et al., 05/02SimulatedActualAutonomous path generationLong Range NavigationRoad Map Navigation (ROAMAN)• Demonstrated onboard path planning algorithm thatautonomously generates a series of waypoints thatare passed to the local path planning algorithm(DriveMaps) for obstacle avoidance during longrange traverses.• Both portions of the algorithm use an occupancygrid representation to perform hazard detection andpath planning. Map pruning leads to highly efficientpath generation.• Maps that are maintained by the higher and lowerlevel portions of the system are not shared, sincethere may be substantial localization errors thataccumulate during any long traverse.• Long range path planning is periodically repeated,depending on camera spatial resolution (typicallygood range data to 12 meters ahead of the rover).Rover Autonomy10T. Huntsberger et al., 05/02Long Range RendezvousRover Autonomy11T. Huntsberger et al., 05/02Lander Detection/Rendezvous•Long range lander tracking/navigation (Fused line andwavelet-derived texture features for target detection,tracking and long-range approach from >100 meters)•Mid range lander tracking/navigation (Multi-linefeature extraction and rover-to-lander pose estimationusing known lander geometry for mid-range approachat 5 – 25 meters)•Lander ramp rendezvous (Pattern extraction,recognition, and precision registered guidance intolander via rover-to-ramp pose estimation on finalapproach at 0.2 – 5 meters)•Continuous-motion mobility: high speed hazarddetection and avoidance for in-route approaches in non-benign terrainRover Autonomy12T. Huntsberger et al., 05/02•Lander localization from long distances (125+meters) using single feature target recognitiontechniques tends to suffer from false positives dueto lack of detail in the lander profile•Autonomous multifeature fusion algorithm usesline features (derived from Canny edge detection)fused with wavelet derived texture signatures toeliminate false positives•Angled line detection gives technique flexibilityfor other navigation operations in close proximity(detection of lander strut structure)• Wavelet derived texture signature allows fastprocessing within rover computing constraints•As illustrated left in a Mars rover to landerapproach, this new technique exploits spatiallocality of the line and texture features for rapidlander localization in the field of viewMultifeature Fusion for Long Range Lander AcquisitionRange: 125 MetersLabeled texturepositivesLabeledlinepositivesFinal fused resultWavelet coefficientspaceRover Autonomy13T. Huntsberger et al., 05/02TargetTargetSingle Command Science Target Rendezvous• Demonstrated single command sequence forautonomous approach to specified remote sciencetargets and instrument arm placement on same inArroyo Seco at JPL.• Science targets were selected from FIDO navcampanorama using WITS running remotely inBuilding 82/PRL at JPL.• 13 step algorithm developed to autonomouslytrack features in vicinity of science targets usingcombination of cross correlation andhomographic transforms.• Relative position of science target updated duringtraverse to mitigate errors in localization usingwheel odometry.Rover Autonomy14T. Huntsberger et al.,