ShepherdingProgressReport (Public)

Transcript

1. Progress Report Distributed Fault-Tolerant Shepherding with Multiple Shepherds Aiyi LI Wakamiya Lab li-aiyi@ist.osaka-u.ac.jp 1 July 9th, 2021
2. Background 2
3. Swarm Intelligence is the collective behavior of decentralized, self-organized systems, natural or artificial Features: Simple rules, no centralized control structure, lead to "intelligent" global behavior Swarm Intelligence 3 https://www.cio.com/article/3192068/why-swarm-intelligence-enhances-business-and-bitcoin.html Examples: ant colonies, bird flocking, animal herding, fish schooling and bacterial growth.
4. Natural shepherding Farmer leads dogs to guide sheep flock towards the goal Characteristics Guiding strategies could vary a lot Dogs receive instructions from farmer Dogs have limited sight Natural shepherding 4 http://www.sundayschoolrevolutionary.com/sunday-school-shepherding/
5. Problem: Shepherds interact with a manageable number of entities, which in turn exercise control over a sheep flock　　　　　 Concept: The shepherding concept provides a biologically inspired framework for multi-agent guidance and human- swarm teaming Shepherding problem 5 References: Abbass, Hussein & Hunjet, Robert. (2021). Shepherding UxVs for Human-Swarm Teaming An Artificial Intelligence Approach to Unmanned X Vehicles: An Artificial Intelligence Approach to Unmanned X Vehicles. 10.1007/978-3-030-60898-9.
6. Models Sheep agent: Boid model Shepherd agent: Combination of forces Shepherd agent operation strategies Rule-based Regard flock as a whole Designate one as target sheep No rule-based Shepherding researches 6
7. Related Works 7 Two examples
8. Regard flock as a whole Collecting or Driving Online target switching 8 References: Strömbom, Daniel, et al. "Solving the shepherding problem: heuristics for herding autonomous, interacting agents." Journal of the royal society interface 11.100 (2014): 20140719. Strategy: Shepherd agent drives when flock gathered, collect when separated
9. Designate one as target sheep Always drive the farthest agent Farthest-agent targeting 9 References: Tsunoda, Yusuke, et al. "Analysis of local-camera-based shepherding navigation." Advanced Robotics 32.23 (2018): 1217-1228. Strategy: Shepherd agent always drive the farthest agent from the goal
10. Two examples Only relates to single shepherd Lots of global factors used Limitation to failures Operation strategies Regard flock as a whole: Failure when separate too much Designate one as target sheep: Available since flock in small scale Analysis 10
11. Our Researches 11
12. Coordination between multiple shepherds No same routes More locale factors used instead of global factors Realistic but efficient Robustness for potential faults Aims 12
13. Omitted Our proposal 13
14. Omitted Our proposal 2th 14
15. Simulations 15
16. One shepherd guides thirty sheep Comparisions 16 Collecting or Driving, 378 steps Drive the Farthest, Failure Our Proposal, 244 Steps
17. Five shepherds guide fifty sheep Comparisions 17 Our Proposal, 423 Steps Collecting or Driving, Failure Drive the Farthest, Failure
18. With the increase of shepherd number Success steps drop correspondingly Test robustness Same tendency of success steps Steps not increased so much with faults Verification 18 when number 3-5, maybe saturated Our proposal, same environment
19. See Gif with fault chance Verification 19
20. Better performance compared to other two methodologies Coordination and robustness at the expectancy Conclusions 20
21. Closing 21
22. Timeline 22 Time Pace January - March Basics & Background April - June Modeling & Programming July Paper writing
23. Shepherd agent operation strategies Rule-based Regard flock as a whole Designate one as target sheep Large-scale simulation No rule-based Dogs find the algorithm itself by machine learning Afterwards 23