Semantic Navigation Maps for Mobile Robot Localization on Planetary Surfaces Gregor Jochmann RIF e.V. Department Robot Technology Joseph-von-Fraunhofer-Str.
Download ReportTranscript Semantic Navigation Maps for Mobile Robot Localization on Planetary Surfaces Gregor Jochmann RIF e.V. Department Robot Technology Joseph-von-Fraunhofer-Str.
Semantic Navigation Maps for Mobile Robot Localization on Planetary Surfaces Gregor Jochmann RIF e.V. Department Robot Technology Joseph-von-Fraunhofer-Str. 20 D-44227 Dortmund Contents ■ ■ ■ ■ ■ Motivation Map Concept Map Generation Map Utilization Applications ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 2 Motivation VEROSIM – The Virtual Robotics Testbed ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 4 VEROSIM – The Virtual Robotics Testbed ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 5 Motivation ■ The Localization Problem model pose sensor data localization [map] [map] ■ Possible application areas ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 6 Map Representation ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 7 Requirements ■ suitable for different ● environments ● localization algorithms ● navigation algorithms ■ consistent, human-comprehensible data representation ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 8 Map Concept Key Aspects ■ landmark-based map ● per landmark □ □ □ □ position orientation (if applicable) type-specific features application-independent quality measure ■ digital elevation model ■ [path network] ■ rules to derive application-specific quality measures Scenarioindependent map Quality measures Map Scenario-specific knowledge ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 10 Plugin Concept ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 11 Map Generation Map Generation Overview Rocks Craters Mountains Shadow segmentation Edge detection Search for DEM maxima Shadow ellipses Edge Pairing Topographical characteristics Landmark detection Crater ellipses Semantic abstraction Perspective projection to global coordinates Semantic abstraction Plausibility tests and cleaning Merging of multiple observations Map generation Valuation of landmarks ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 13 Cleaning ■ ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 14 Merging ■ ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 15 Valuation ■ ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 16 Map Utilization Landmark Identification ■ mapping: observations vs. landmarks in map ● ● ● ● requirement for localization positional information landmark-specific features robot pose information ■ benefit from of semantic information: ● typed landmarks ● distance measure feature weights ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 18 Localization Algorithms – Particle Filter ■ ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 19 Localization Algorithms – Kalman Filters ■ ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 20 Applications Planetary Environments ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 22 Forest Environments ■ driver assistance ■ tree documentation ■ tree detection with laser scanners and stereo cameras ■ additional sensors ■ simulated and real data ASTRA 2013, Noordwijk, The Netherlands. Gregor Jochmann, RIF, Department Robot Technology 23