New Assignment Portion for Versatile Robots - PowerPoint PPT Presentation

emergent task allocation for mobile robots l.
Skip this Video
Loading SlideShow in 5 Seconds..
New Assignment Portion for Versatile Robots PowerPoint Presentation
New Assignment Portion for Versatile Robots

play fullscreen
1 / 36
Download Presentation
Download Presentation

New Assignment Portion for Versatile Robots

Presentation Transcript

  1. Emergent Task Allocation for Mobile Robots Nuzhet Atay Doctoral Student Seminar Advisor : Burchan Bayazit

  2. Motivation • Given an unknown environment and a swarm of mobile robots • Achieve some goals under a set of constraints • Explore the environment • Regions of interest • Dynamic • Unpredictable • Spread or shrink • Obstacles • Real-life scenarios Nuzhet Atay

  3. Robotic Systems • Heterogeneous robots with limited • Speed • Sensing range • Communication range • Multiple robot coordination • Task allocation • Goal: • Optimum assignment of robots Nuzhet Atay

  4. Planning and Task Allocation Task Distribution Task Distribution • Multi-robot systems require efficient and accurate planning • Global optimal solutions are expensive • communication overhead • planning time • Our Solution: An emergent approach • Emergent: Solution results from interactions of robots • Local approximation to global optimal • Low cost and feasible in real-time Nuzhet Atay

  5. Outline • Problem Definition • Model • Centralized (Global Optimal) Solution • Emergent Approach • Comparison of two methods • Experimental Results • Conclusion Nuzhet Atay

  6. Problem Definition • Objective is to assign robots to • Cover regions of interest • Provide communication between all robots • Control maximum total surface • Explore new regions • We can define this problem as an optimization problem • Given the robot information and environment properties • What is each robot’s ideal next step? Nuzhet Atay

  7. Model • Robots • Constant communication and sensing range • Limited speed • Regions of interest • Targets that need to be tracked by the robots • Several robots may be needed • Input: • Information about the robots and the environment • Expected target positions after n steps • Output • Optimum locations of robots Nuzhet Atay

  8. Centralized Solution Task Assignment Information Collection • Problem is defined as a mixed-integer linear program • Non-linear constraints • Flexible • Easy to customize • Objective: maximize • Target Coverage • Communication • Area Coverage • Exploration R7 T3 R8 R9 R6 T1 R4 R5 R3 T2 R1 R2 Central Server Task Allocation is Determined Nuzhet Atay

  9. Target Coverage • Each robot has a sensing range • Each target has a coverage requirement • A target is covered • Necessary number of robots has the target in sensing range Nuzhet Atay

  10. Communication • Two robots can communicate • If they are within communication range of each other • There is a series of robots that can provide communication Nuzhet Atay

  11. Area Coverage • Maximum area coverage is obtained • Sensor overlap is minimized Nuzhet Atay

  12. Exploration • Robots store the places they have visited • Each robot tries to locate itself outside the explored region Nuzhet Atay

  13. Optimum Distribution • Sample distribution for maximizing • Target coverage • Communication • Area coverage Nuzhet Atay

  14. Additional Constraints • Environment obstacles • Convex • Partitioned into convex obstacles • Convex box surrounding the obstacle Nuzhet Atay

  15. Problems of Global Approach • Solution is not feasible with large number of robots • Solving mixed-integer linear program is NP-Hard • Central server • Too much data transfer • Our solution: • Solve small local problems • Integrate to approximate optimal solution • Advantage is to avoid • Communication overhead • Exponential computation time Nuzhet Atay

  16. Emergent Task Allocation Find a Solution with Local Information Information Sharing • Robots find optimal solutions with local information • Each robot has different information about • Robots in the environment • Targets to be tracked • Environment properties • Solutions are different • Independent suboptimal solutions • To find better solution, robots • Exchange information • Recompute their solutions • Final result depends on • Information content • Number of iterations T3 R7 R8 R9 R6 T1 R4 R5 R3 T2 R1 R2 Recompute Solution with Updated Information Nuzhet Atay

  17. Intentions • Robots send their intended locations to neighbors • Each robot assumes these locations are final • Finds its optimal location Nuzhet Atay

  18. Directives • Robots send expected locations of neighbors • Each robot chooses the best among them Nuzhet Atay

  19. Intentions and Directives • Robots send both their and neighbors computed locations • Each robot finds the best location using options Nuzhet Atay

  20. Intentions, Directives and Target • Robots send • Their and their neighbors’s locations • Possible target assignments • Each robot • Decides a target assignment • Finds the best location using options Nuzhet Atay

  21. Comparison of Global and Emergent • Emergent approach is more efficient • Computation • Communication • Approximate solution to global optimal Nuzhet Atay

  22. Experiments • How well emergent performs? • Comparison to global • Experiment scenario • 8 robots • 6 targets • 3 obstacles • How scalable is the emergent? • 20 robots – 10 targets • 30 robots – 15 targets Nuzhet Atay

  23. Solution - Global Nuzhet Atay

  24. Solution - Emergent Nuzhet Atay

  25. Comparison • Solution quality is comparable # of Targets Covered at Each Step Nuzhet Atay

  26. Evaluation • Emergent approach is 400 times faster than global approach Solution Time Nuzhet Atay

  27. Scalability • Execution time remains constant with larger networks Nuzhet Atay

  28. Conclusion • Planning framework for multi-robot task allocation • Low communication cost and suitable for real-time applications • 400 times faster than the global optimal solution • Comparable solution • Future work: • Different negotiation strategies • Implementation on real robots • Different utility functions Nuzhet Atay

  29. Questions Motion Planning Group ? Nuzhet Atay

  30. Convergence • ETA approaches to CGO after finite number of steps • Observation: • If all robots find the same solution, then this solution is the same as CGO • At each step • Robots find a solution • Exchange information and negotiate • Assuming all state information is shared • Robots will have information about other robots’ views • After p steps • All robots have the same information and find the same solution Nuzhet Atay

  31. Solution Quality Nuzhet Atay

  32. Solution Quality Nuzhet Atay

  33. Solution Quality Nuzhet Atay

  34. Solution Quality Nuzhet Atay

  35. Solution Quality Nuzhet Atay

  36. Solution Quality Nuzhet Atay