1. Fail-Safe Mobility Management and Collision Prevention Platform for Cooperative Mobile Robots with Asynchronous Communications Rami Yared School of Information Science Japan Advanced Institute of Science and Technology (JAIST) Supervised by: Prof. Xavier Défago 1

2. Application Garden Cultivation by cooperative mobile robots. 2

3. Context Group of mobile robots Asynchronous communication (No upper bound on communication delays) No upper bounds on robots speeds No central control 3

4. Problem Prevent collisions between mobile robots. 4

5. Research Objective Mobility management platform Fail-safe mobile robotic system Prevent robots collisions. 5

6. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 6

7. 7 Motion planning Find a route from an initial position to a final position in presence of obstacles.

8. Related work Avoid collision between a robot and Fixed obstacles Sensing during the motion in dynamic or unknown environments 8 Minguez et al 2004. [22] Montano et al 1997. [23] Motion planning RT guarantees

9. Related work Upper bound on communication delays. Upper bound on processing speeds. Wireless LAN, Access point central router 9 Synchronous systems Nett et al 2003 [25]

10. Related work 10 Synchronous systems Nett et al 2003 [25] Collisions between mobile robots Violation of timeliness properties

11. Related work Time elastic: Time bounds can be increased or decreased dynamically Fail safe: exhibits correct behavior, or put the system in a fail-safe state. 11 Martins et al 2005 [21]

12. Related work 12 Martins et al 2005 [21] Collisions between mobile robots

13. Wireless Communications ⇒ retransmission mechanisms. Arbitrary sized messages ⇒ unknown delays, not anticipated,... ⇒ Time free approach is important 13

14. Contribution Time free mobility management platform Fail-Safe mobile robotic system. Collision prevention protocols: Closed group of robots. Dynamic group of robots. 14

15. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 15

16. 16 Motion planning Find a route from an initial position to a final position in presence of obstacles.

17. System architecture 17 Fail-safe Time free

18. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 18

19. System model Asynchronous communications Retransmission ⇒ reliable channels Positioning system with bounded errors. 19

20. Approach Distributed path reservation system. Primitives: Request Reserve Release 20

21. Reserve / Release 21

22. Specification Safety A given zone can be owned by only one robot. Zone i ∩ Zone j ≠ ∅ ⇒ (R i owns Zone i ) XOR (R j owns Zone j ) 22

23. Specification Liveness If R i requests Zone i then eventually (R i owns Zone i or an Exception is raised) R i requests Zone i ⇒  (R i owns Zone i or Exception) 23

24. Specification Raising exceptions occurs only in specified situations. Non triviality Exception is raised only if a deadlock situation occurs. 24

25. 25 Reserved Zone ε gps : Positioning system ε tr : translation movement ε θ : rotation movement

26. Request / Released zone 26

27. Deadlock situation 27 Deadlock situation Robot R i requests a resource owned by R j Robot R j requests a resource owned by R i

28. Starvation situation 28 Starvation situation If robot R j owns Zone j then R i is blocked (starvation) Pathological situation

29. 29 Next Zone j RiRi

30. 30 Next Zone j Deadlock situation

31. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 31

32. Part 1: Collision prevention protocol for a closed group of mobile robots. 32

33. Closed group model Composition known to all robots Communication graph is fully connected 33

34. Collision prevention protocol Requests ordering wait-for relations between robots Consistency All robots agrees on the same wait-for relations. 34

35. Total Order Broadcast 35 TO-broadcast TO-deliver

36. Protocol 36 When Request() Compute the requested zone TO-broadcast(Request, Zone, Release previous zone) When TO-deliver(Request, Z, Release previous zone) update the wait-for graph Dag wait When vertex becomes a sink (no outgoing edges) Reserve zone

37. Example 37

38. Fault-tolerant collision prevention 38 Robots fail by crash Communication part Total Order Broadcast Problem: If a robot has crashed A robot waiting for a crashed robot is blocked The number of blocked robots increases ⇒ Snowball effect A robot cannot distinguish a crashed robot from a very slow one (asynchronous system) Zone d Zone j Zone b Zone i Zone a

39. Fault-tolerant collision prevention 39 Robots fail by crash with a failure detector class P with a failure detector class ♢ P with a failure detector class ♢ S Solution: Zone d Zone j Zone b Zone i Zone a

40. Fault-tolerant collision prevention 40 Robots fail by crash with a failure detector class P Perfect failure detector The suspected robot is considered as an inert obstacle A waiting robot becomes unblocked. Solution: Zone d Zone j Zone b Zone i Zone a

41. Fault-tolerant collision prevention 41 Robots fail by crash with a failure detector class ♢ P Eventually perfect failure detector Preemptive protocol Solution: Zone d Zone j Zone b Zone i Zone a

42. Fault-tolerant collision prevention 42 Preemptive protocol If a robot R d is suspected then Zone d is “blocked” Requests of R a and R j are preempted (alternative zones) Other robots R i and R b are not blocked. Zone d Zone j Zone b Zone i Zone a

43. Fault-tolerant collision prevention 43 Preemptive protocol If a robot R i is suspected and has not owned Zone i then Request of R i is preempted (restarts its request of Zone i ) Robot R b is not blocked. Zone b Zone i

44. Fault-tolerant collision prevention 44 with a failure detector class ♢ S Non preemptive protocol If R i suspects R j and Zone i intersects with Zone j then R i cancels its request of Zone i (alternative zone) Zone j Zone i

45. Fault-tolerant collision prevention 45 Failure detector class ♢ P Liveness property for the preemptive protocol, because eventually a correct robot is not suspected by any correct robot. Failure detector class ♢ S Liveness property for the non preemptive protocol. Requires more alternative zones.

46. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 46

47. Part 2: Collision prevention protocol for a dynamic group of mobile robots. 47

48. Dynamic group model 48 limited transmission range, No routing is required Communication graph is not connected

49. Reservation range 49 Reservation range ≤ Transmission range / 2 D ch ≤ D tr / 2

50. Input of Neighborhood Discovery: (x,y) coordinates of the caller. Output of Neighborhood Discovery: the set of robots that potentially conflict with the caller. Neighborhood discovery 50

51. Ngh i = {R a, R b, R d, R e, R j } G i = {R b, R j } (G1) i = {R b } (G2) i = {R j } WLAfter i = {R k } Collision prevention protocol 51

52. Collision prevention protocol 52

53. Performance Analysis Robots are active executing the protocol reservation range (D ch) density of robots (s) Average effective speed vs reservation range Average effective speed vs density of robots 53

54. Performance Analysis Average communication delays T com Delay of the neighborhood discovery primitive T nd Physical speed of robots V mot Average effective speed V 54

55. Performance Analysis 55

56. Performance Analysis Effective speed vs reservation range. range 56

57. Effective speed vs density of robots Performance Analysis 57

58. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 58

59. Conclusion 59 Closed groupDynamic group group of robots StaticDynamic group knowledge Completepartial Scalability (design) Lowvery high Fault-tolerance ♢S♢S

60. Closed groupDynamic group messages loss Safety violation Imprecision positioning system Safety violation Neighborhood discovery Safety violation 60 Conclusion Vulnerability with respect to system model assumptions

61. Outline Related work and motivation System architecture System model and problem specification Fail-safe platform Collision prevention for a closed group model Collision prevention for a dynamic group model Conclusion Future directions 61

62. Future directions 62 Simulation Optimizations

63. Thank you for your attention 63