Multi-Modal Localization for Autonomous Lunar Lander Robert Fisher - PowerPoint PPT Presentation

Multi-Modal Localization for Autonomous Lunar Lander Robert Fisher Heather Jones

(NOT TO SCALE) Fairing First Stage Separation Separation De-orbit and Braking Launch Descent Trans- Lunar Injection Earth Orbit Lunar Orbit Lunar Orbit Insertion Second Stage Separation

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 3

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 4

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 5

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 6

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 7

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar 8

Sensors • Inertial Measurement Unit • Star Tracker • Radio Telemetry • Radar Altimeter • Point Lidar(s) • Downlooking Camera(s) • Flash Lidar

Dataset • Lunar Reconnaissance Orbiter • Imagery (with latitude/longitude bounding box) • Laser altimeter data (with latitude/longitude coordinates of each point) • Digital Elevation Map of the entire moon constructed from laser altimeter data (118m x 118m per pixel)

Multiple modes of data for one solution • (Blum, Mitchell ’98) introduced ‘co - training’ for semi-supervised learning. • Co-training principles apply in a variety of applications which are not semi-supervised

When to use multiple views? • We need to be able to localize using each view independently. • Each localizer needs to give a comparable confidence measure (i.e. variance of the distributions)

When to use multiple views?

Relation to previous work • Previous work (Singh, Lim, 2008) showed that using altimetry combined with imagery performs better than either mode alone. • We wish to investigate additional modes, perhaps multiple different views extracted from image data.

References • Avrim Blum, Tom Mitchell. “Combining labeled and unlabeled data with co- training”, COLT 1998. • Leena Singh, Sungyung Lim. “On Lunar on - orbit Vision-Based Navigation: Terrain Mapping, Feature Tracking driven EKF”, AIAA Guidance 2008.