Evolutionary algorithms for robot path planning, task allocation and collision avoidance in an automated warehouse

Croucamp, Marco (2022-04)

Thesis (PhD)--Stellenbosch University, 2022.

Thesis

ENGLISH ABSTRACT: Research with regard to path planning, task allocation and collision avoidance is important for improving the field of warehouse automation. The dissertation addresses the topic of routing warehouse picking and binning robots. The purpose of this dissertation is to develop a single objective and multi-objective algorithm framework that can sequence products to be picked or binned, allocate the products to robots and optimise the routing through the warehouse. The sequence of the picking and binning tasks ultimately determines the total time for picking and binning all of the parts. The objectives of the algorithm framework are to minimise the total time for travelling as well as the total time idling, given the number of robots available to perform the picking and binning functions. The algorithm framework incorporates collision avoidance since the aisle width does not allow two robots to pass each other. The routing problem sets the foundation for solving the sequencing and allocation problem. The best heuristic from the routing problem is used as the strategy for routing the robots in the sequencing and allocation problem. The routing heuristics used to test the framework in this dissertation include the return heuristic, the s-shape heuristic, the midpoint heuristic and the largest gap heuristic. The metaheuristic solution strategies for single objective part sequencing and allocating problem include the covariance matrix adaptation evolution strategy (CMA-ES) algorithm, the genetic algorithm (GA), the guaranteed convergence particle swarm optimisation (GCPSO) algorithm, and the self-adaptive differential evolution algorithm with neighbourhood search (SaNSDE). The evolutionary multi-objective algorithms considered in this dissertation are the non-dominated sorting genetic algorithm III (NSGA-III), the multi-objective evolutionary algorithm based on decomposition (MOEAD), the multiple objective particle swarm optimisation (MOPSO), and the multi-objective covariance matrix adaptation evolution strategy (MO-CMA-ES). Solving the robot routing problem showed that the return routing heuristic outperformed the s-shape, largest gap and midpoint heuristics with a significant margin. The return heuristic was thus used for solving the routing of robots in the part sequencing and allocation problem. The framework was able to create feasible real-world solutions for the part sequencing and allocation problem. The results from the single objective problem showed that the CMA-ES algorithm outperformed the other metaheuristics on the part sequencing and allocation problem. The second best performing metaheuristic was the SaNSDE. The GA was the third best metaheuristic and the worst performing metaheuristic was the GCPSO. The multi-objective framework was able to produce feasible trade-off solutions and MOPSO was shown to be the best EMO algorithm to use for accuracy. If a large spread and number of Pareto solutions are the most important concern, MOEAD should be used. The research contributions include the incorporation of collision avoidance in the robot routing problem when using single and multi-objective algorithms as solution strategies. This dissertation contributes to the research relating to the performance of metaheuristics and evolutionary multi-objective algorithms on routing, sequencing, and allocation problems. To the best of the author’s knowledge, this dissertation is the first where these four metaheuristics and evolutionary multi-objective algorithms have been tested for solving the robot picking and binning problem, given that all collisions must be avoided. It is also the first time that this specific variation of the part sequencing and allocation problem has been solved using metaheuristics and evolutionary multi-objective algorithms, taking into account that all collisions must be avoided.

AFRIKAANSE OPSOMMING: Navorsing in verband met roete beplanning, part allokasie en botsing vermyding is belangrik vir die bevordering van die pakhuis automatisering veld. Die verhandeling handel oor die onderwerp van parte wat gestoor en gehaal moet word en die verkillende parte moet ook gealokeer word aan ’n spesifieke robot. Die doel van hierdie verhandeling is om ’n enkele doelwit en ’n multidoelwit algoritme raamwerk te ontwikkel wat parte in ’n volgorde rangskik en ook die parte aan ’n robot alokeer. Die roete wat die robot moet volg deur die pakhuis moet ook geoptimeer word om die minste tyd te verg. Die volgorde van die parte bepaal uiteindelik die totale tyd wat dit neem vir die robot om al die parte te stoor en te gaan haal. Die doelwitte van die algoritme raamwerk is om die totale reistyd en die totale ledige tyd te minimeer, gegewe die aantal beskikbare robotte in die sisteem om die stoor en gaan haal funksies uit te voer. Die algoritme raamwerk bevat botsingsvermyding, aangesien die gangbreedte van die pakhuis nie toelaat dat twee robotte mekaar kan verbygaan nie. Die roete probleem lˆe die grondslag vir die oplossing van die volgorde en allokerings probleem. Die beste heuristiek vir die roete probleem word verder gebruik in die volgorde en allokerings probleem. Die verskillende roete heuristieke wat in hierdie verhandeling oorweeg was, sluit in die terugkeer heuristiek, die s-vorm heuristiek, die middelpunt heuristiek en die grootste gaping heuristiek. Die metaheuristieke vir die volgorde en allokerings probleem sluit die volgende algoritmes in: die kovariansie matriks aanpassing evolusie algoritme (CMA-ES), die genetiese algoritme (GA), die gewaarborgde konvergerende deeltjie swermoptimerings (GCPSO) algoritme, en laastens die selfaanpassende differensi¨ele evolusie algoritme met die teenwoordigheid van buurtsoek (SaNSDE). Die evolusionêre multidoelwit algoritmes wat oorweeg was vir die volgorde en allokerings probleem sluit die volgende algoritmes in: die multidoelwit kovariansie matriks aanpassing evolusie algoritme (MO-CMA-ES), die nie-dominerende sortering genetiese algoritme III (NSGA-III), die multidoelwit evolusionˆere algoritme gebaseer op ontbinding (MOEAD) en laastens die multidoelwit deeltjie swermoptimering algoritme (MOPSO) Oplossings van die robot roete probleem het gewys dat die terugkeer heuristiek die s-vorm, grootste gaping en middelpunt heuristiek met ’n beduidende marge oortref het. Die terugkeer heuristiek is dus gebruik vir die oplossing van die roete beplanning van robotte in die volgorde en allokasie probleem. Die raamwerk was doeltreffend en die resultate het getoon, vir die enkel doelwit probleem, dat die CMA-ES algoritme beter gevaar het as die ander metaheuristieke vir die volgorde en allokasie probleem. Die SaNSDE was die naas beste presterende metaheuristiek. Die GA was die derde beste metaheuristiek, en die metaheuristiek wat die slegste gevaar het, was die GCPSO. Vir die multidoelwit probleem het die MOPSO die beste gevaar, as akkuraatheid die belangrikste doelwit is. As ’n grootter verskeidenheid die belangrikste is, is die MOEAD meer geskik om ’n oplossing te vind. Die navorsingsbydraes sluit in dat vermyding van botsings in ag geneem word in die robot roete probleem. Hierdie verhandeling dra by tot die navorsing oor die oplossing van roete beplanning, volgorde en allokasie probleme met metaheuristieke. Na die beste van die outeur se kennis is hierdie die eerste keer dat al vier metaheuristieke getoets was om die robot stoor-en-gaan haal probleem op te los, onder die kondisie dat alle botsings vermy moet word. Dit is ook die eerste keer dat hierdie spesifieke variant, enkel-en-multidoelwit probleem van die volgorde en allokasie van parte met behulp van metaheuristieke en multidoelwit evolusionˆere algoritmes opgelos was, met die inagneming dat alle botsings vermy moet word.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/124515
This item appears in the following collections: