A Tale of Two Rovers Mission Scenarios for

A Tale of Two Rovers: Mission Scenarios for Kilometer-Scale Site Survey David Thompson and David Wettergreen The Field Robotics Center, Carnegie Mellon

Agenda Scarab: Surveying lunar regolith in polar craters Mission scenario Mobility Navigation and localization Zoë: Intelligent surficial mapping Feature detection and classification Kilometer-scale adaptive site survey Conclusions

Terrain Difficulty Technologies for site survey Dark Navigation Robust Localization Slope and Crater Mobility Science Autonomy Integration with Orbital Data Autonomous Data Collection Autonomous Traverse Distance

Scarab: lunar mission scenario Land in crater Supervised autonomy (polar orbiter relay) Site Survey of regolith composition, hydrogen content 7 months, 25 drill sites over 25 kilometers

Mobility requirements Challenging terrain extreme slopes loose soil Navigation and localization in lunaranalog environments 5. 0 cm/s dark navigation

A stable science / drilling platform 270 kg mass 250 kg to counter drill thrust low CG high torque 100 kg Science payload 1 m coring drill Dark Navigation Sensors Regolith Drill Core System Hazard Avoidance Sensors Science Payload “Differencing” Linkage Radioisotope Generator Simulator Avionics Body Raise/Lower Linkage & Actuator

Suspension Skid steering Passive terrain matching / body averaging Actuated rocker arms permit leveling / drilling on slopes Kneels during drilling operations improve stability maximize drilling depth / minimize wasted travel

Auto-leveling

Inchworming Theoretical benefits max slope: 19 vs. 23 degrees drawbar pull: 1038 vs 1281 N Conventional rolling

Inchworming

Navigation Dark navigation with active sensing Laser light striping Laser scan merging (courtesy NASA ARC) Traversability analysis, D* path planning

NEPTEC Tri. DAR Raster resolution to 512 x 512 30 -degree FOV accurate geologic maps for drill site selection

Localization Wheel odometery is unreliable Kalman-filtered IMU 3 -axis ring laser gyro 3 -axis acceleration Optical velocity sensor with ground lighting

Field Tests 10 h 50 m, 1090 m (2. 8 cm/s) June – Mobility and autonomy testing at Moses lake WA November – Science payload tests in Hawaii

Agenda Scarab: Surveying lunar regolith in polar craters Mission scenario Mobility Navigation and localization Zoë: Intelligent surficial mapping Feature detection and classification Kilometer-scale adaptive site survey Conclusions

Zoë: Surficial Survey Mission image courtesy Dom Jonak, CMU Multiple-kilometer autonomous traverses 1 m/s continuous travel in open terrain Autonomous science feature recognition, data collection, and mapping Tests at Amboy Crater, Mojave desert, CA

Autonomous VISNIR acquisition Automatic rock detection Wide-baseline stereo estimates rock position Autonomous spectrum classification

MVJ detector for variable lighting candidate bounding boxes h 1 h 2 hn . . . rock bounding boxes nonrock cascade 1 h 2 nonrock hn . . . cascade 2. . . input image h 1 h 2 . . . hn nonrock cascade m max

Rock detection and visual servo SIFT matching recognizes and tracks dozens of targets Science-relevant maps Permits visual servo

Spectrum acquisition

Tracking performance Detection and tracking: 21 (± 3. 9) rock spectra in 40 min Blind pointing: 0 rock spectra 50 40 Lost Track Miss 30 Miss 20 10 Rock Spectra Lost Track Miss Rock Spectra

Rock detection N Rock Detection Precision: 90. 8% (± 2. 6, � =0. 05) run 4 rocks run 3 3. 0 2. 0 run 2 final rover position run 1 1. 0 0. 0

Adaptive surficial mapping

Adaptive surficial mapping “Gaussian process” terrain model Site survey informed by surface and orbital data Maximum-entropy sampling chooses optimal observation sites

Inference Result 450 m autonomous traverse Extrapolates by interpreting orbital images Discovers map parameters on the fly

Informative path planning Science-driven Adds robustness to execution uncertainty

Recovery from Navigation Error

Fidelity of Reconstructed Maps reconstruction accuracy 0. 9 0. 8 Previously reported at i. SAIRAS 2008 – Thompson, Wettergreen 0. 7 0. 6 0. 5 0 50 100 150 200 250 number of returned features Fixed, transect: Fixed, coverage pattern: Adaptive, low-res orbital: Adaptive, high-res orbital: 74% 75% 81% 87% 300 (± 0. 09) (± 0. 05) (± 0. 03) (± 0. 01)

Conclusions Mobility improvements facilitate new operational modes involving kilometerscale site survey Future work Selective data return (image analysis and spatial statistics) Data fusion for science and navigation (DEMs, orbital and surface images)

Thanks! Field Robotics Center: David Wettergreen, Red Whittaker, David Kohanbash, Paul Bartlett, Dom Jonak, Jason Zigler Johnson, Glenn, NORCAT, ARC Scarab: NASA Human-Robot Systems research program, grants NNX 08 -AJ 99 G (Robert Ambrose) and NNX 07 -AE 30 G (John Caruso). Zoë: NASA ASTEP NNG 04 GB 66 G (David Lavery)