Improving Orchard Efficiency with Autonomous Utility Vehicles
Download as PPTX, PDF1 like1,097 views
The document discusses the use of autonomous utility vehicles to improve efficiency in orchard maintenance tasks by enabling multiple functions such as mowing and spraying. It describes the technical implementation of these vehicles using laser range sensors and the Hough transform for detecting tree rows, facilitating navigation and task execution. Ongoing efforts aim to enhance performance in uneven terrains and improve automation for specialty crop management.
Improving Orchard Efficiency with Autonomous Utility Vehicles
- 1. Improving Orchard Efficiency with Autonomous Utility VehiclesBrad Hamner, Sanjiv Singh, and Marcel BergermanCarnegie Mellon University
- 2. Orchard Tasks TodayMaintenance tasks require humans to drive vehicles a large distance over a long time
- 3. Reconfigurable MobilityOne type of vehicle can perform multiple tasksMowing, spraying, etc.In use more oftenOne automation package can be applied to multiple vehiclesOnly one system to learn
- 4. Autonomous Prime MoverAutomated Toro eWorkmanDemonstrate orchard tasksby towing otherequipment
- 5. Autonomous Prime MoverLaser range sensorsFind treesFind obstaclesSteering angle sensorDistance traveled sensorControlled by laptop
- 6. Seeing the Trees, Finding the RowFrom laser sensors,robot sees one slice of the worldNeed to find tree rows from this dataTwo parallel lines
- 7. Seeing the Trees, Finding the RowFrom laser sensors,robot sees one slice of the worldNeed to find tree rows from this dataTwo parallel lines
- 8. Line RepresentationLines represented by their perpendicularto the vehicledistance d, angle qTree pointqd
- 9. Hough Transform Row DetectionHough Transform – Popular line detector in image processingCreate “votes” for linesLooking for parallel lines with most votes
- 10. Hough TransformPick a point in the image How about this one?
- 11. Hough TransformMany lines could go through that point.Draw them, and record their d,qAngleDistance
- 12. Hough TransformPick another point. Draw its vote lines,and record their d,qAngleDistance
- 13. Hough TransformPick another point and repeat.AngleDistance
- 15. Hough TransformRepeat for all points.Two lines dominate the voting.AngleDistance
- 16. Hough TransformFind the top two peaks in the matrix.Those are the tree rows.600Actual Hough Matrix500400Angle300200100DistanceVotes
- 17. Hough TransformFind the top two peaks in the matrix.Those are the tree rows.600Actual Hough Matrix500400Angle300200100DistanceVotes
- 18. Row FollowingHough matrix peaks are the tree rowsDrive midway between the two
- 19. TurningWe know roughly how wide rows arePlan a path into the next rowAs the vehicle gets close, run row detection
- 21. ResultsOver 100 km autonomous travelOrchards in PA and WADemonstrationsMowingSprayingCrop scouting
- 22. Results10 km with an autonomous platform
- 23. Ongoing WorkUneven terrain brings the ground into viewPresents data that looks like tree rowsConfuses Hough detector
- 24. Ongoing WorkParticle filterTracks possible rows over timeRejects bad rowsMore stable resultsParticleFilterHoughalone
- 25. Ongoing WorkNot always much room to turn aroundDetect the available spacePlan smooth trajectoriesBack up if necessary
- 26. Related CollaborationComprehensive Automation for Specialty Cropshttp://www.frc.ri.cmu.edu/casc/CASC/Welcome.htmlInexpensive localizationImprove tracking, build mapsInformation managementAutonomous disease detectionExtension and usabilityUser interfaceCommercialization studiesCost benefit analysis
- 27. AcknowledgmentsUSDA Specialty Crop Research InitiativeOwners and managers ofSoergel Orchards Penn State Fruit Research and Extension Center Bear Mountain OrchardsHollabaugh Orchards Sunrise Orchards Valley Fruit Orchards
Editor's Notes
- #3: Inefficient use of a person’s time