1. TECHLAV Annual Meeting Greensboro, NC May 31-June 1, 2017
Testing, Evaluation, and Control of Heterogeneous
Large-Scale Systems of Autonomous Vehicles (TECHLAV)
TECHLAV Annual Meeting
Greensboro, NC May 31-June 1, 2017

2. Task Allocation Using Parallelized Clustering and Auctioning Algorithms for Heterogeneous Robotic Swarms Operating on a Cloud Network
By: Jonathan Lwowski, Dr. Patrick Benavidez, Dr. Jeff Prevost, and Dr. Mo Jamshidi

3. Overview Introduction System Overview Clustering Auctioning
Genetic Algorithm
Hardware Emulation Results
Conclusion

4. Motivation Costa Concordia cruise ship crash
45250 passengers
33 deaths and 64 injuries
Goals of Task T1-6 (Cloud-Based Control of LSASV) and Sub-thrust 1-2 (Cooperative Localization, Navigation and Control of LSASV)
Heterogenous swarm robotics
Task allocation
Navigation/Path Planning
Autonomous Cloud-based robotics
Data simplification and organization

5. Significance of Methods
One of the first heterogenous swarms of MAVs and ASVs to use a cloud-based network to simplify and organize data to be used for task allocation
Parallelization of data simplification algorithms on the cloud to increase speed and scalability
Tested in both software simulations and hardware emulations

6. System Overview

7. Localizing the People MAVs equipped with bottom facing stereo cameras
People represented by colored circles in simulation
People detected by simple color thresholding techniques
People localized using stereo camera model and coordinate transformations

8. Clustering the People Used to simplify the map
Group people into clusters of a given radius
Constrained Cluster Radius K-mean Clustering
Parallelized CCR K-means Clustering on the cloud

9. Clustering the People
500 people and a desired cluster radius of 10 meters

10. Parallelization of Clustering Algorithm

11. Meta-clustering the Clusters
Groups the clusters into meta-clusters
Number of meta-clusters same as the number of ASVs
Fuzzy C-means Clustering
Allows clusters to be assigned to multiple meta-clusters

12. Meta-clustering the Clusters
500 people, desired cluster radius of 10 meters, 5 ASVs

13. Auctioning the Meta-clusters
Used to assign a meta-cluster to each ASV
Uses ASVs locations, capacities, and speeds
Uses meta-clusters locations and number of people in the meta-cluster
Uses mothership location

14. Auctioning the Meta-clusters

15. Traveling to Assigned Clusters
Genetic algorithm
Fitness value = 1/distance
50 random permutations chosen as initial population

16. Hardware Emulation Results
GPS emulation
Overhead webcam
Ar-track-alvar
Provides x,y, and z position of each Ar-Tag
Used to emulate the GPS and stereo camera localization of the MAVs

17. Hardware Emulation Results
GPS Emulation

18. Hardware Emulation Results
CCR K-mean Clustering
10 Ar-Tags representing people and 4 Clusters

19. Hardware Emulation Results
CCR K-mean Clustering
10 Ar-Tags representing people and 4 Clusters

20. Hardware Emulation Results
Video

21. Conclusion
One of the first heterogenous swarms of MAVs and ASVs to use a cloud based network to simplify and organize the data to be used for task allocation
Parallelization of data simplification algorithms to increase speed and scalability of the system
Effectiveness of systems demonstrated in software and hardware experiments

22. Future Work Use deep learning to detect and localize the people
Add autonomous underwater robots to detect people under the surface of the water
During the summer, Dr. Joordens will implement this using model boats on an actual body of water

23. References
J. Lwowski, P. Benavidez, J. J. Prevost, and M. Jamshidi, "Task Allocation Using Parallelized Clustering and Auctioning Algorithms for Heterogeneous Robotic Swarms Operating on a Cloud Network,” Chapter 3 in Autonomy and Artificial Intelligence, (W. F. Lawless, et al, eds.), Springer-Verlag, 2017 (In Press)