- Open Access
A new multiple robot path planning algorithm: dynamic distributed particle swarm optimization
© The Author(s) 2017
- Received: 21 September 2017
- Accepted: 14 October 2017
- Published: 2 November 2017
Multiple robot systems have become a major study concern in the field of robotic research. Their control becomes unreliable and even infeasible if the number of robots increases. In this paper, a new dynamic distributed particle swarm optimization (D2PSO) algorithm is proposed for trajectory path planning of multiple robots in order to find collision-free optimal path for each robot in the environment. The proposed approach consists in calculating two local optima detectors, LODpBest and LODgBest. Particles which are unable to improve their personal best and global best for predefined number of successive iterations would be replaced with restructured ones. Stagnation and local optima problems would be avoided by adding diversity to the population, without losing the fast convergence characteristic of PSO. Experiments with multiple robots are provided and proved effectiveness of such approach compared with the distributed PSO.
- Multiple robots
- Path planning
- Local optimum detectors
The concept of multiple robot systems (MRS) began in the 1990s, in particular in works regrouping mobile robots, gathering objects  and robot colonies [2, 3]. Arai et al.  identified seven primary research themes in the MRS: biological inspirations, communication, architectures, location/cartography/exploration, transport and handling of objects, motion coordination and reconfigurable robots.
Multiple robot systems are well known by the synchronization process and having better spatial distribution capability as compared to a single robot. This coordination addresses the problem of how teams of autonomous mobile robots can share the same workspace while avoiding interference with each other, collision with static obstacles and/or while achieving group motion objectives.
There are two basic approaches to solve the problem of multiple robot path planning: centralized and distributed. In the case of the centralized approach, each robot is treated as a composite system, and planning is done in a composite configuration space, formed by combining the configuration spaces of the individual robots. While in the case of the distributed approach, paths are first generated for robots independently and then their interactions are considered. The advantage of centralized approaches is that they always find a solution when there is one. However, the practical difficulty is the temporal complexity which is exponential in composite configuration space. Distributed planners help generate robot trajectories independently before using different strategies to resolve potential conflicts. But, they are incomplete in nature (probabilities of various and varied configurations) and can therefore lead to blocking situations. This distributed approach can be applied to each robot taking into account the positions and orientations of all other robots at each point in time. Thus, the general problem is reduced to several versions of the planning problem for a single mobile robot in the presence of other robots which move in the presence of fixed obstacles. A trajectory of the robot is then found by the search of a path from the start to the arrival thanks to the spatial–temporal configuration.
The control of the MRS becomes unreliable and even infeasible if the number of robots increases. In addition, the multiple robot path planning problem becomes more and more complex. The latter has been extensively studied since the 1980s. Swarm behavior has proven its effectiveness in such problems thanks to interesting properties like robustness, flexibility and scalability. One of the successful optimization methods is particle swarm optimization (PSO) algorithm.
This paper proposed a novel approach to determine the optimal trajectory of the path for distributed multiple robots system using dynamic distributed particle swarm optimization (D2PSO), where each robot is considered to be a mobile, autonomous and physically independent agent.
The remaining part of the paper is outlined as follows. “Literature survey of particle swarm optimization use in MRS path planning” section covers briefly the latest works done in the MRS path planning search domain using different PSO variants. Formulation of the problem for multiple robot path planning has been elaborated in “Problem formulation for multiple robot navigation” section. “Obstacle avoidance approach” section describes our obstacle avoidance approach. The classical particle swarm optimization, dynamic distributed double guided particle swarm optimization algorithm and dynamic distributed particle swarm optimization are described in the “Particle swarm optimization (PSO) for MRS path planning” section. “Conclusion” section demonstrates the computer simulation for path planning of multiple robots.
Since the inception of PSO [5, 6], several variants have been proposed to improve the performance of original PSO. The first versions of PSO for MRS were proposed in [7–9] to find a target in a given environment, and studies have demonstrated that the PSO algorithm has acceptable performances in the searching task. In the study of Chakraborty et al. , behavioral cooperation of the robots was realized through selection of alternative local trajectories for collision avoidance among teammates. In fact, he compared the performances using differential evolution (DE) with a PSO-based realization.
The authors present in  PSO-based technique for determining the optimal set of parameters for a second PSO for collective robotic search. Particle swarm optimization technique was used to optimize the velocity parameters of robots in , to arrive at the shortest collision-free trajectory, satisfying dynamic constraints. A hybrid technique for the control of swarms of robots, based on particle swarm optimization (PSO) and consensus algorithms, is presented in . A MOPSO algorithm is utilized in  to generate trajectories for mobile robots that are working on the environments that the robots are working on and may be included danger sources. Darvishzadeh and Bhanu  present a framework to use a modified PSO (MPSO) algorithm in a multiple robot system for search task in real-world environments. Nakisa et al.  also proposes a new method (APSO) to create an efficient balance between exploration and exploitation by hybridizing basic PSO algorithm with A-star algorithm. Nakisa proposes a method based on the multi-swarm particle swarm optimization (PSO) with local search on the multiple robot search system to find a given target in a complex environment that contains static obstacles . Rastgoo et al.  proposed an algorithm named the “modified PSO with local search (ML-PSO)” applied in the exploration search space by adding a local search algorithm such as A-star to guarantee global convergence with a reduction in the search time. Allawi and Abdalla  used PSO combined with reciprocal velocity obstacles (RVO) method, in order to choose the best paths for robots without collision and to get to their goals faster. Das  proposed a new methodology to determine the optimal trajectory of the path for multiple robot in a clutter environment using hybridization of improved particle swarm optimization (IPSO) with an improved gravitational search algorithm (IGSA).
A hybridization of improved particle swarm optimization (IPSO) with differentially perturbed velocity (DV) algorithm (IPSO-DV) was also proposed by Das et al.  for trajectory path planning of multiple robots in a static environment. Abbas et al. discusses in  an optimal path planning algorithm based on an adaptive multi-objective particle swarm optimization algorithm (AMOPSO) for five robots to reach the shortest path. The algorithm PSO-NAV presented in Raffaele Grandi’s work  focuses on the possibility to drive a group of very simple robots from a starting zone to a final one inside a maze-like environment unknown a priori.
The obstacle avoidance approaches in MRS studies aim to find a path from an initial position S of a robot to a desired goal position G, with respect to positions and shapes of known obstacles O. The penalty function to be minimized by the planning algorithm consists of two parts. While the first one evaluates a length of the trajectory (or time needed to execute the trajectory), the second part ensures safety of the path (i.e., distance to obstacles).
The origin of this method comes from the observations made during computer simulations on grouped flights of birds and fish . These simulations highlighted the ability of individuals in a moving group to maintain an optimal distance between each other’s and to follow a global movement in relation to neighbors one.
To apply the PSO, we have to define a particle search space and an objective function to optimize. The principle of the algorithm is to move these particles so that they find the optimum.
A position, that is, its coordinates in the definition set.
A speed that allows the particle to move. In this way, during the iterations, each particle changes its position. It evolves according to its best neighbor, its best position and its previous position. This evolution makes it possible to fall on an optimal particle.
A neighborhood, that is, a set of particles that interact directly with the particle, especially the one with the best criteria.
Its best position visited. The value of the calculated criterion and its coordinates are essentially retained.
The position of the best neighbor of the swarm that corresponds to the optimal scheduling.
The value of the objective functions because it is necessary to compare the value of the criterion given by the current particle with the optimal value.
PSO is initialized with a group of random particles (solutions) and then searches for optima by updating generations. In every iteration, each particle is updated by following two “best” values. The first one is the best solution (fitness) the particle has achieved so far. This value is called pBest. Another “best” value that is tracked by the particle swarm optimizer is the best value obtained so far by any particle in the population. This best value is a global best and called gBest.
D3GPSO: the dynamic distributed double guided particle swarm optimization algorithm
The D3GPSO introduced by Bouamama in [4, 27] is a distributed PSO. It is a group of agents dynamically created and cooperating in order to solve a problem. Each agent performs locally its own PSO algorithm.
Inspired by works in [1, 2, 4, 28], this algorithm uses the same principle as the D3G2A , and it consists on dividing the initial population into subpopulations and affecting each one to an agent. Each agent is also called specie agent and is responsible of a set of particles having their fitness values in the same range. This range is called FVR (for fitness value range), and it is the specificity of the specie agent SpecieFVR. The species agents cooperate all by exchanging solutions to reach the optimal one. In fact, each one executes its own double guided PSO algorithm. The latter is double guided by the concept of template and the min-conflict heuristic. It is enhanced by new parameters: guidance probability P guid; a local optimum detector LOD and a weight ε (used by species agents to calculate their own PSO parameters).
For the D3GPSO, we distinguish also a mediator agent to manage the communication between the species agents. This agent, called interface agent, can also create new species agents if necessary.
The local optimum detector LOD is an operator that we use in the PSO process. It represents the number of iterations in which the neighboring does not give improvement. If the best solution found by a specie agent remains unchanged for LOD generations, we can conclude that the particles are blocked in a local optimum. So, the best particle having this fitness value will be penalized. This variable is given by the user at the beginning of the optimization process, but it is changed by each specie agent according to the attained fitness value.
D2PSO: dynamic distributed PSO for MRS path planning
In 2006, Hereford  has introduced a version of the PSO that “distributes” the motion processing among several, simple, compact, mobile robots, called distributed PSO (dPSO). Calculations were done “locally,” that is on each local robot. Simulation results showed that the dPSO appears to be a very good way of coordinating simple robots for a searching task operation. One of the most important advantages was that the algorithm appears to be scalable to large numbers of robots since the communication requirements do not increase as the number of robots is increased.
Although swarm intelligence approaches are attractive methods for robotic target searching problems, these strategies have two important disadvantages: First, they may get stuck on local optima. Second, they have slow progress in terms of fitness function in some situations (slow speed to converge to the target locations).
Inspiring from the D3GPSO described in [4, 27], we introduce two new parameters to the PSO: local optima detector for global best LODgBest and local optima detector for personal best LODpBest. The purpose of the latest parameter is to count the number of successive iterations for which personal best and global best do not give improvement. Since these particles are unable to improve their pBest, they are no more contributing in finding the global optimal solution. This indicates that particles are saturated and require external thrust to boost their power. Dynamic distributed PSO (D2PSO) provides thrust by heading particles toward potentially better unexplored regions which also add diversity to the search space. At the same time, when global best gBest is not improving for predefined number of successive iteration, it may be trapped in local optima and mislead other particles by attracting toward it. This also requires some external push that send trapped particle outside local optima position and mitigate its consequences. By this way, the stagnation and local optima problems would be avoided without losing the fast convergence characteristic of PSO since the D2PSO would follow the PSO’s behavior for the rest of situations.
Local optimum detector
LOD is a parameter to the whole optimization process, and it will be locally and dynamically updated by each robot [1, 2]. If the personal best of the ith particle increases for a specific number of successive generations, we can conclude that the particle optimization sub-process is trapped in a local optimum, and so the LODpBest will increment, respectively, for gBest and LODgBest.
Our work consists in implementing a dynamic distributed version of PSO. It is a multi-agent approach. It acquires its dynamic aspect from the agents that it uses. Indeed, they are capable to calculate their own parameters on the basis of the parameters given by the user (S p , S g , ε). ε ϵ [0, 1].
As mentioned in , we define in (8) and (9) the parameters α and β which are function of templatepBest and templategBest. Their worth is that they articulate the probability for a particle to propagate its knowledge. This confirms the fact that the best particles have more chance to be pursued by others.
The multiple robot path planning algorithm is implemented in a simulated environment. The simulation is conducted through programming in C language on a Pentium microprocessor, and robot is represented with similar soft-bots of rectangular shape with different path color code. The robot is self-contained in terms of power. It is mobile; it may be limited in terms of steering radius and speed, but it is mobile.
Predefined initial location and goal location for all the robots are assigned. The experiments were conducted with three different radius obstacles and assigned same velocities for each robot at the time of the program run.
Initial and target points used in simulation
We have applied the distributed PSO and the proposed D2PSO to the same environment. Population size M was fixed to 150 and then to 300.
We used the following parameters: inertia weight w = 1.0, damping ratio w damp = 0.99, personal learning coefficient c 1 = 1.5, global learning coefficient c 2 = 1.5, S p = 3, S g = 3.
Experiment results and analysis
We made several simulation runs. We have run the program for 3, 5, 7 and 10 robots with different initial and target points, for 200 iterations. We evaluated the effectiveness of the algorithm by comparing the path lengths obtained from the basic distributed PSO and the D2PSO, ensuring of course the non-collision with the static predefined obstacles (Figure 3).
Path lengths for M = 150
M = 150
Path lengths for M = 300
M = 300
Since restructuring both pBest and gBest depends on the number of robots and the population size, we deduce that each time we increase the number of bots, we obtain better values of the objective function, and respectively when increasing the population size M.
Execution time for M = 150
M = 150
Execution time for M = 300
M = 300
Comparison of number of iterations taken to obtain the shortest path by the different algorithms
The problem of multiple robot motion planning focuses on computation of paths of different robots such that each robot has an optimal path, and so the overall path of all the robots combined is optimal. Many approaches have been proposed for solving multiple robot path planning problems. Particle swarm optimization algorithm is one of the successful optimization methods in this area. This paper has presented a successful improvement to the PSO algorithm. D2PSO ensures diversity to stagnated particles in such a manner that they move to better and unexplored regions of search space. In addition, it does not disturb the fast convergence characteristics of PSO by keeping the basic concept of PSO unaffected. Experimental results show that our approach performs better in escaping local optimum and proves that applying D2PSO to multiple robots path planning problem is practical and efficient for large number of robots in environments with variable obstacles.
The main contributions of our research are: (1) finding optimal paths of mobile robots moving together in the same workspace, (2) proposing to use the PSO evolutionary algorithm and (3) ensuring collision-free trajectories.
However, there are still some issues and improvements to be addressed in our future work. First, dynamic obstacles, unknown environment, obstacles’ shapes and collision avoidance should be studied. In this paper, both the environment and obstacles are static relative to the robots, which is applicable in particular cases. In the future, work will be carried out using dynamic obstacles during the multiple robot path planning process. Second, the inter-robot collision should be considered in future experiments. The task planning process for MRS would be also studied in order to ensure best coordination.
AA formulated, analyzed and implemented the D2PSO and algorithms described in the paper inspiring from D3GPSO. SB provided guidance and valuable suggestions for the improvement in the paper. Both authors read and approved the final manuscript.
I would like to thank the editor and the anonymous references for their helpful comments.
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Bouammama S, Ghedira K. Une nouvelle génération d’algorithmes génétiques guidés distribués pour la résolution des Max_CSPs. Revue des sciences et technologie de l’information, serie Technique et science informatique, TSI. 2008;27(1–2):109–40.Google Scholar
- Bouammama S, Ghedira K. A dynamic distributed double guided genetic algorithm for optimization and constraint reasoning. Int J Comput Intell Res. 2006;2(2):181–90.View ArticleGoogle Scholar
- Arai T, Pagello E, Parker LE. Editorial, “Advances in multi-robot systems”. IEEE Trans Robot Autom. 2002;18(5):655–61.View ArticleGoogle Scholar
- Bouamama S, Ghedira K. A family of distributed double guided genetic algorithm for Max_CSPs. Int J Knowl Based Intell Eng Syst. 2006;10(5):363–76.View ArticleGoogle Scholar
- Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of the IEEE international conference neural networks, Perth, Australia, 1995, vol. 4, p. 1942–8.Google Scholar
- Eberhart RC, Kennedy J. A new optimizer using particle swarm theory. In: Proceeding of the 6th international symposium micromachine human science, Nagoya, Japan, 1995, p. 39–43.Google Scholar
- Doctor S, Venayagamoorthy G, Gudise V. Optimal PSO for collective robotic search applications. In: IEEE congress on evolutionary computation, Portland, OR, June 2004, p. 1390–5.Google Scholar
- Pugh J, Segapelli L, Martinoli A. Applying aspects of multi-robot search to particle swarm optimization. In: International workshop on ant colony optimization and swarm intelligence, Brussels, Belgium, 2006, p. 506–7.Google Scholar
- Hereford JM. A distributed particle swarm optimization algorithm for swarm robotic applications. In: IEEE congress on evolutionary computation, Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada, July 16–21, 2006.Google Scholar
- Chakraborty J, Konar A, Chakrabortyand UK, Jain LC. Distributed cooperative multi-robot path planning using differential evolution. In: IEEE world congress on computational intelligence, 2008.Google Scholar
- Venayagamoorthy GK et al. Optimal PSO for collective robotic search applications. In: Proceedings of the congress on evolutionary computation, 2004. CEC2004, Institute of Electrical and Electronics Engineers (IEEE), Jan 2004.Google Scholar
- Grandi R, Falconi R, Melchiorri C. Coordination and control of autonomous mobile robot groups using a hybrid technique based on particle swarm optimization and consensus. In: ROBIO conference, 2013.Google Scholar
- Zhang Y, Gong D-W, Zhang J-H. Robot path planning in uncertain environment using multi-objective particle swarm optimization. Neurocomputing. 2013;103:172–85.View ArticleGoogle Scholar
- Darvishzadeh A, Bhanu B. Distributed multi-robot search in the real-world using modified particle swarm optimization. In: GECCO’14, July 12–16, 2014, Vancouver, BC, Canada.Google Scholar
- Nakisa B, Rastgoo MN, Norodin MdJ. Balancing exploration and exploitation in particle swarm optimization on search tasking research. J Appl Sci Eng Technol. 2014;8:1429–34.Google Scholar
- Grandi R, Falconi R, Melchiorri C. A particle swarm optimization-based multi robot navigation strategy. In: International workshop on bio-inspired robots, 2011.Google Scholar
- Rastgoo MN, Nakisa B, Zakree M, Nazri A. A hybrid of modified PSO and local search on a multi-robot search system”. Int J Adv Robot Syst. 2015;12:86.View ArticleGoogle Scholar
- Allawi ZT, Abdalla TY. A PSO-optimized reciprocal velocity obstacles algorithm for navigation of multiple mobile robots. Int J Robot Autom (IJRA). 2015;4(1):31–40.Google Scholar
- Das PK, Behera HS, Panigrahi BK. A hybridization of an improved particle swarm optimization and gravitational search algorithm for multi-robot path planning. Swarm Evol Comput. 2016;28:14–28.View ArticleGoogle Scholar
- Das PK, Behera HS, Das S, Tripathy HK, Panigrahi BK, Pradhan SK. A hybrid improved PSO-DV algorithm for multi robot path planning in a clutter environment. Neurocomputing. 2016;207(C):735–53.View ArticleGoogle Scholar
- Abbas NH, Abdulsaheb JA. An adaptive multi-objective particle swarm optimization algorithm for multi-robot path planning. J Eng. 2016;22(7):164–81.Google Scholar
- Nakisa B, Rastgoo MN, Nasrudin MF, Zakree M, Nazri A. A multi-swarm particle swarm optimization with local search on multi-robot search system. J Theoret Appl Inf Technol. 2015;71:129–36.Google Scholar
- Latombe JC. Robot motion planning. Norwell: Kluwer; 1991.View ArticleMATHGoogle Scholar
- Borenstein J, Koren Y. The vector field histogram: fast obstacle avoidance for mobile robots. IEEE J Robot Autom. 1991;7(3):278–88.View ArticleGoogle Scholar
- Meng H, Picton PD. A neural network for collision-free path planning. Artif Neural Netw. 1992;2(1):591–4.Google Scholar
- Reynolds CW. A distributed behavioral model. In: Proceedings of the 14th annual conference on computer graphics and interactive techniques, 1987, vol. 21, issue 4, p. 25–34.Google Scholar
- Bouamama S. A new distributed particle swarm optimization algorithm for constraint reasoning. In: Proceedings of the 14th international conference on knowledge-based and intelligent information and engineering systems: part II, Sept 08–10, Cardiff, UK, 2010.Google Scholar
- Hu X, Eberhart R. Solving constrained nonlinear optimization problems with particle swarm optimization. In: 6th world multiconference on systemics, cybernetics and informatics, Orlando, Florida, USA, 2002.Google Scholar
- Deneubourg JL, Beckers R, Holland OE. From local actions to global tasks: stigmergy and collective robotics. In: Proceedings of the fourth international workshop on the synthesis and simulation of living systems, Cambridge, MA, USA, July 1994, p. 181–9.Google Scholar