Page - 37 - in Joint Austrian Computer Vision and Robotics Workshop 2020

only in a very small amount. For this reason, and sinceourgoalwith thedeclarativeencoding is to im- prove the overall quality of the assignment, in the ASP implementation we minimize the overall trav- elling distance, like we do for the task assignment. The rules and constraints needed are very similar to the ones we used before, where PR1, PR2 are some user-definedparameters required for theassignment: Listing 2ASP encoding of the charge assignment - assignment rule 0{charge (S,R) : stat ion (S,PR1,PR2)}1 :− robot (R,PR1,PR2, ) . 3.2.TheIncubedITUseCase IncubedIT isa roboticscompanyfocusedonsoft- ware development for smart robots. They typically deal with problems of the same type as the previ- oususe-casewe justdiscussedabove. Thus themain topicismulti-robotplanningandscheduling. Forthis reason,programmersat IncubedITdesignedahighly parameterized platform which, if configured accord- ingly, can facea lotofdifferent situations, likeware- houses of online traders, logistic centers of super- markets and car manufacturing plants. Fortunately, this platform is quite modular, partially centralized and partially decentralized, with a main FMS mod- ule which is responsible for the coordination of the many parts of the system. Thanks to this design, re- placing the old solving module with the ASP solver hasbeeneasy todo. In the imperative implementation, two kinds of optimization costs can be used: FIFO and global op- timum. The former does not require more explana- tions,while the latterconsidersaprioritynumberas- sociatedtoeachtask. Regardingthetaskassignment, we stick to the important constraint rule in ASP: we can assign only one vehicle to a task, and only one task to a vehicle at a time. The same rules and con- straintsweused for theBMWuse-case thusfit to In- cubed ITsoftwareaswell. We can now focus on the other problem to solve, the charge assignment. The charging strategy here is more sophisticated than in the BMW-case, and robot canbesent tocharge for fourdifferent reasons: fixed time slot charging: robots are assigned to charging stations due to a reached time slot; critical charg- ing: robots are assigned to charging stations due to a battery level below the critical charging limit; busy charging: robots are assigned to charging stations due to a battery level below the busy charge limit; idle charging: robots are assigned to a charging sta- tion due to not enough appropriate assignable tasks. Obviously, all of these parameters (critical and busy charging limit, duration of the time slot) can be cus- tomized by the user. We define now the rules and constraintsused to implement the third situation: Listing3Assignmentofbusycharging robots 0{charge (S,R busy) : chargingstation (S , , , , ) , robot stat ion (R,S)}1 :− robot charge opt (R,BL, automatic mode , , , ,BCL,CCL) , BL <= BCL, BL > CCL. Listing4Avoidanceofdoubleallocations :− charge (S,R, ) , charge (S,R2, ) ,R!=R2. :− charge (S,R, ) , charge (S2 ,R, ) ,S!=S2. :− assign (T,R) , charge ( ,R, ) . Incontrast to theBMWcase-study,herewedonot handle the two problems of task and charge assign- ment separately: weoptimize twodifferentweighted criteria. The most important one is the minimization of the overall travelling distance of robots assigned to tasks or to charging stations due to forced time- slot, critical or busy charging. Then, the same op- timization, with a lower weight, is applied to robots assigned to parking places and charging stations for idlecharging. 4. Evaluation of Runtime and Quality for bothCase-Studies In this section, we present a brief evaluation of both case-studies. We designed several instances for each case-study involving different numbers of robots, orders, charging and parking stations to test different scales. Subsequent, the runtime as well as the quality of the solutions for these scenarios are compared. Furthermore, since Clingo can combine different meta-heuristics and parallelization strate- gies for the solving process, we tested all the com- binations between them in order to find, for each case, the best one. As a result of the evaluation of these solving approaches [6], we chose the branch- and-bound-based optimization strategy in combina- tion with splitting-based search multithreading and four threads for the two BMW assignment prob- lems, while for IncubedIT the best approach is the Vsids Heuristic combined with compete-based mul- tithreadingwith four threads. 37