A Decentralised Strategy gy for r Hete teroge geneous AUV Mis issions via ia Goal l Dis istribution and Temporal Pla lanning Yaniel Carreno, Èric Pairet, Yvan Petillot, and Ron Petrick Edinburgh Centre for Robotics Heriot-Watt University & University of Edinburgh Edinburgh, Scotland, United Kingdom {y.carreno, eric.pairet, y.r.petillot, r.petrick}@hw.ac.uk edinburgh-robotics.org | orcahub.org
Robotics in in th the Underwater Domain Lab Environment Simulation Environment • Long-te term de deplo loymen ents ts in in th the unde derwate ter domain in usually lly req equir ire th the in intr troductio ion of of multi ti-robot syste tems as as serv ervic ice-orie riente ted age agent nts. • Multi lti-robot syste tems can support complex missio ions th that over ercome many of of th the lim limita itatio tions of of sin ingle le-robot solu lutio tions, pro rovid iding ing ro robustn tness to to the he overa verall ll syste tem. edinburgh-robotics.org | orcahub.org
Temporal Pla lanning for Hete teroge geneous Robots • In In re real-world rld applic icatio ions of of robotic ics, planning ing pro rovid ides solu lutio ions in in enablin ling autonomous and in intellig lligent beh ehavio iours for hete hetero rogen eneous robotic ic platfo form rms to to co comple lete its ts ta tasks. • Robotic tic pla latfo forms: missio ion dep deplo loyments with th fleet eets of of Aut Autonomous Und nderwater Veh Vehic icle les (AU (AUVs). • Ap Approach: model as as tem temporal pla lanning ing pro roble lems. Conc ncurr rrent t act ctio ion n exe executio ion with th di diffe fere rent t ti time e sl slots. ‒ Make ke use use of nu numeric ric co constra rain ints. ‒ Use se tem tempora ral l ref eferenc nces to to so solv lve mul ulti-vehic icle le pro roble lems. ‒ Al Allo low th the e imple lementatio tion of co coordin inated ac actio ions. ‒ • Bui Build ld on on previous approache hes that that use use tem tempora ral pla lannin ning (e (e.g., PAN PANDORA pro roje ject & EU EUROPtus). Sing Si ngle le robot t approach. ‒ Requir equire e fast pla lannin ing g tec techn hniq iques (s (small ll pla lannin ning ti times) and nd (ne (near) r) optim imal l goa goal l di dist strib ributio ion. ‒ Often ten poor ta task allo locatio tion in n com comple lex missio ions with th larg rge nu numbers rs of goa goals. ‒ edinburgh-robotics.org | orcahub.org
Experimental Results – task allocation(1) Ben Benchm hmark rk planne anners ge gene nerate rate non non-optim timal l goa goal di distr trib ibutio tions (left ft – PO POPF, ri righ ght – OP OPTIC) in n the he unde underwater scena nario rio. Poor Poor goa goal di distrib tributio ion n af affe fects missio ion costs: : en energy ergy consumptio ion and and total tal di distan tance e trave ravelled lled. edinburgh-robotics.org | orcahub.org
Task Allocation Strategy PL PLAN ANNER AGN AGNOSTIC TIC APPR APPROACH OUT OUTPUT edinburgh-robotics.org | orcahub.org
Task Allocation Strategy PL PLAN ANNER AGN AGNOSTIC TIC APPR APPROACH CAP APABILI ILITY ALLO ALLOCATIO ION OUT OUTPUT edinburgh-robotics.org | orcahub.org
Task Allocation Strategy PL PLAN ANNER AGN AGNOSTIC TIC APPR APPROACH ROB OBOTS DI DIST STRIB IBUT UTIO ION OUT OUTPUT edinburgh-robotics.org | orcahub.org
Task Allocation Strategy PL PLAN ANNER AGN AGNOSTIC TIC APPR APPROACH GO GOALS LS ALLO ALLOCATIO TION OUT OUTPUT edinburgh-robotics.org | orcahub.org
System Framework The The DHRTA module outp tput is is a set of of mission goals ls describ ibed in in sta tanda dard PDDL with ith th the follo llowin ing structure: (goal_ al_na name param aramete ters)[r [robot] t] DE DECENTRALISED PL PLANNING MIS ISSION EX EXECUTION TASK K ALL ALLOCATION edinburgh-robotics.org | orcahub.org
DHRTA & Temporal Planning Mis ission Defi finit ition (a (auv1): The Decentralis lised Het eterogeneous Robots Task All llocator (DHRTA) module le allo llocates mis ission go goals ls to to a PDDL Domain in: PDDL Prob oble lem: fleet of of heterogen geneous AU AUVs and dis istri ributes the go goals ls (define (domain auvs_inspection) (define (problem task) (:requirements …) (:domain auvs_inspection) in in a dece decent ntralis lised manne ner. (:types …) (:objects …) (:predicates …) (:init …) (:functions …) (:goal (and (:durative- action navigation …) (poi_valve_turned off poi55))) (:durative- action broadcast_data …) (:durative- action rock_inspection …) (:metric minimize (total-time)) (:durative- action valve_inspection …) ) ROS OSPla lan n (auv auv1) (:durative- action valve_turning …) … OPTI PTIC C Pla lanner PLA LAN PLAN DISPATCH PLA MI MISSI SSION GOAL ALS TAS ASK DIST STRIBUTION (poi_valve_turned off poi55) (poi_ i_valve_ e_tu turn rned ed off poi5 i55) 5)[auv uv1] 1] (poi_valve_turned on poi35) (poi_ i_valve_ e_tu turn rned ed on poi3 i35) 5)[auv uv2] 2] Miss ssio ion n Def efin init itio ion (a (auvN) (poi_rock_analysis poi40) (poi_ i_ro rock_analysis is poi4 i40) 0)[auv uv2] 2] (poi_soil_analysis poi34)[auv2] … (poi_soil_analysis poi34) … ROSP SPla lan (a (auvN) edinburgh-robotics.org | orcahub.org
Experimental Results – task allocation(2) • DHR DHRTA+TP improve ves goal al di distr trib ibutio tion n and and overa verall ll comple lexity ity of the he global l plan an. • Op Optim timis ises the he use use of robots and and resou esources. edinburgh-robotics.org | orcahub.org
DHRTA-TP: Experimental Results • DHR HRTA-TP TP gen genera rates the the first so solv lvable le pla lan in in sho short rter ti time perio eriod than than ben enchmark pla lanners rs (l (left eft to top). • The The makespan resu esult lts (l (lef eft bottom) are si simila ilar to to the the value lues obtain ined by by the the benc enchmark pla lann nners. • DHRTA-TP TP strategy outperf rforms the he cen entralis lised approach, achie ievin ing the he hig ighest number of of goals ls du durin ing the the miss ssion for all goa goal set sets. edinburgh-robotics.org | orcahub.org
Thanks!! Yaniel Carreno Èric Pairet Yvan Petillot Ron Petrick edinburgh-robotics.org | orcahub.org
Recommend
More recommend
