Towards Context-Aware Navigation for Long-Term Autonomy in - PowerPoint PPT Presentation

Towards Context-Aware Navigation for Long-Term Autonomy in Agricultural Environments A U TH ORS : M A R K H ÖL L M A NN ∗ , BENJ A MI N K I S L IUK ∗ , J A N C H R I STOP H K R A U SE ∗ , C H R I S TOPH TI EB EN ∗ , A L EX A NDER M OCK †, S EBA STIAN P Ü TZ †, F EL I X I G EL B R IN K †, TH OM A S W I EM A NN ∗ †, S A NTIAG O F OCK E M A RTI NEZ ∗ , S TEFA N S TI EN E ∗ , J OA CH I M H ER TZ BERG ∗ † S P EA KER : BENJ A M IN K I S LIU K ∗ D F K I L A B L OW ER S A X ONY, P L A N B A S ED R OB OT C ON TR OL † U NI V ERSI TY OF OS NA BRÜCK, K NOW L EDG E BA S ED S Y S TEMS

Motivation The future of agriculture is already a central driver for robotic innovation • Integrated long term autonomy benefits robotic applications • Farms are non-standardized work environments with large, heterogeneous areas • Robots need to work in and switch between different contexts 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 2

Prior and Related Work Navigation: • Eband, DWA, Global Planner, Move Base Flex, Localization: • Robot localization, AMCL Control Architecture: • SMACH Environment Modelling & Representation: • Waypoint Server, Costmap 2D, QGis 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 3

Experimental Setup 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 4

Experimental Setup Autonomous Robotic Experimental Platform: • RTK Multi-GNSS • WIFI & 4G Data uplink • Multiple 2D and 3D LIDAR sensors Base Station Container: • WIFI & 4G Data Uplink • Inductive Charging Station • Monitoring and Data Server 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 5

Context Aware Navigation Different contexts in different areas: • Area usage / crop type • Time of day • Season • State of an agricultural process Topological planning: • Abstract Path Planning • Each zone represents a context • Different set of parameters for each context 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 6

Context Aware Navigation Different contexts in different areas: • Area usage / crop type • Time of day • Season • State of an agricultural process Topological planning: • Abstract Path Planning • Each zone represents a context • Different set of parameters for each context 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 7

Context Aware Navigation Different contexts in different areas: • Area usage / crop type • Time of day • Season • State of an agricultural process Topological planning: • Abstract Path Planning • Each zone represents a context • Different set of parameters for each context 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 8

Context Aware Navigation 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 9

Context Aware Navigation • Goal and robot pose are added into the graph as temporary vertices • These vertices are connected with all waypoints in the same zone • Dijkstra's algorithm to compute path in the graph • Each edge traversed is a path segment • Move base flex is called to execute the segment with parameters given from the zone model 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 10

Context Aware Navigation • Goal and robot pose are added into the graph as temporary vertices • These vertices are connected with all waypoints in the same zone • Dijkstra's algorithm to compute path in the graph • Each edge traversed is a path segment • Move base flex is called to execute the segment with parameters given from the zone model 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 11

Context Aware Navigation 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 12

Video 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 13

Future Work – System Enhancements Further extend Move Base Flex: • Adapt costmap representation • More specialized controllers • Automated zone generation From Contexts to Semantics: • Enrich Zones with semantically inferable information • Generate Zones with logical reasoning • Adapt behaviour based on semantic knowledge 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 14

Conclusion • Integration of multiple heterogenous contexts is required to achieve long term autonomy • Hierarchical abstraction with flexible behaviour implementation facilitates the adaption of robotic behaviour according to semantic inference • The flexible architecture allows different state-of-the-art software stacks to be applied where most beneficial 14.09.2020 IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 15

Acknowledgements Sponsored by the Work by the University of Osnabrück and DFKI is supported by the German Federal Ministry of Education and Research in the project SoilAssist2 (Grant No. 031B0684D) and by the German Federal Ministry of Food and Agriculture within the experimental field Agro-Nordwest project (Grant No. 28DE103E18) and by the Ministry of Science and Culture of Lower Saxony and within the Zukunftslabor Agrar (11-76251-14-3/19 (ZN3490) IROS 2020 -TOWARDS CONTEXT AWARE NAVIGATION – DFKI & UNI OSNABRUECK 16 14.09.2020