Swarm Robots Crossing Small Voids:
Practical Considerations
Niken Syafitri
Electronics and Computer Engineering University of Southampton
Southampton, UK
Richard M. Crowder Electronics and Computer Engineering
University of Southampton Southampton, UK
Paul H. Chappell
Electronics and Computer Engineering University of Southampton
Southampton, UK
Abstract—This paper discusses a strategy for a robotic swarm, consisting of small low cost mobile robots, to cross a void or crevice in the environment. To cross the void, the swarm robots are required to autonomously construct a structure. The individual swarm robots have no knowledge of the swarm size, the void location or its width. In addition all robots have an equal probability to locate the void and initiate the structures construction. Once all robots have crossed the void, they disas- semble and continue with the task. If the swarm determines that there is not enough robots to reach the other side of the void while maintaining the structure’s stability, the structure will retreat and the robots disperse. This paper discusses the algorithm required to cross the void and the resulting simulation results, together the practical challenges of implementation such a system using low cost mobile robotic platforms.
I. INTRODUCTION
Social insects can perform sophisticated tasks when work- ing collaboratively, under unpredictable conditions, which are impossible if they work independently, for example ants crossing voids, bees and wasps building their nests, and termites building nests with complex internal structure [1], [2].
Taking the inspiration from nature, especially ant colonies, the adaptivity of real-world systems to overcome unstructured and unknown environment can be achieved through the application of carefully designed algorithms and mechatronic. This is reflected in the developed of robotic swarms. The characteris- tics of a robotic swam include (1) autonomous mobility; (2) limited sensing and communication abilities; (3) simplicity;
(4) decentralized control and coordination mechanism; (5) homogeneity and (6) scalability [3]. This paper considers the development of an approach that allows a robotic swarm to autonomously cross a void, in particular the algorithm and the robots design specification.
While solutions have proposed heterogeneity to solve the unstructured environment problem [4], [5], it can be costly as a range of robots are required and each type should be present in large numbers. Related approaches to problems such as climbing obstacles [6], [7], [8], [9], and avoiding holes or traverse a void [10], [11], [12], [13] by assembling swarm robots into a structure or robotic organism have been proposed.
A complimentary approach was reported in [14] discusses the development of a vertical self-assembled structure that mimics the pyramidal tower structure of robots. However, many of the reported proposed approaches employ the strategy
of assembling all robots into structure, after which the whole structure climbs, avoids or crosses a void. After crossing the obstacle, the structure disassembles. Unfortunately, this method has significant problems when the size of the void changes or is unknown. In addition a strategy must be provided if the robot swarm cannot cross the obstacle, to maintain the integrity of the robotic swarm.
From a swarm robotics perspective, research has fo- cused on control and application [15], [16], [17], mostly achieved through simulation. This can leave a lack of un- derstanding about exactly how real-time interconnections of robots to form structures is possible. From the current self- assembling/reconfiguring modular swarm robotic systems, the approaches for interconnecting units vary quite significantly.
This is because almost every swarm robotics platform or modular robot has significant differences in sensing, communi- cation and actuation, hence each robot family requires a unique interconnection ability. It is important to note that docking recognition and inter-robot communication are as significant as the physical approaches to docking. The robots need to be able to interconnect and release intelligently.
In order to maintain the structural integrity and the rigidity of the structure, we have placed a limit on the size of the void that can be crossed by a linear simple structure to be no more than the equivalent of five robots wide. If the void is wider that this, a far more complex structure is required, the design and implementation of which is outside the scope of this paper.
The paper is structured as follows, the required strategy is presented in Section II, simulation results are discussed in Section II-B. Section III considers the design constraints for a small mobile robot capable of crossing a small void. The paper concludes with Section IV.
II. STRATEGY FORVOIDCROSSING
In developing the approach to crossing a void by au- tonomously constructing a self supporting structure, a number of broad assumptions have been made regarding the robot and its operational environment, namely;
• The surface on the environment is smooth.
• The elevation of both sides of the void are identical, the sides of the void are vertical.
• The robot platform being used is based on our previously developed low cost Formica robot [18]. The upgraded
(a) The robot that locates the void halts at the edge and waits until two addition robots couple to it.
(b) Once three robots are present, the structure advances by one robot over the void.
Fig. 1. Key steps in construction the structure.
version is approximately 8cm × 8cm and weights ap- proximately 200gm
We considered the structure required to cross the void and concluded that for voids less than five robots wide a structure one robot wide would be satisfactory, this was based on consideration of the proposed couplings discussed in Section III-B. In the simulations the physical docking process was not considered, in addition to the challenges of obtaining and maintains the alignment between individual robots, and the robot structure and the edge of the void.
A. Strategy
In the simulation a simple environment is used that consists of two areas in which the swarm robots and wander, separated by the void. The robots are initially placed in the start area, and cross the void to the finisharea.
The proposed strategy is as follows: the robots initially conducts a random exploration of the start area until one robot detects the void. Once this robot detects the void, it becomes theseedrobot and broadcasts a message to all members of the swarm in range, to initiate the construction of the structure.
The rear robot of the structure calls the closest robot to joint the structure until the number of robots in the structure passes a preset parameter, normally three, Figure 1. Then the structure advances across the void and continue calling other wandering robots to join to maintain the structural stability as the structure advances. When a part of the structure has crossed the void and the number of robots at the finish area again satisfies the structural stability, the front robot detaches itself from the structure and returns to a wandering mode. The process of robots joining the structure continues, while the structure continually moves forwards, until there is no robots left to join the structure. The individual swarm robots, have no prior knowledge of the total number in the swarm, the void’s location and or its width. In addition any swarm member must be able to become the seed robot to initiate construction.
In the case when the number of robots available is not enough to cross the void, or if the structural stability criteria is not satisfied, the structure will retreat, and the robots will disassemble from the structure and return to re-exploring the start area. This part of scenario is to prevent the loss or damage to the robots or part of the swarm so that the swarm can accomplish other tasks.
Fig. 2. The number of robots required in the structure to cross a three robot wide void.
The strategy of small void crossing is centered on maintain- ing the correct number of robots in the structure, to ensure that the structures remains stable. Figure 2 shows the construction of a structure to cross a three robot wide gap. If there are N robots in the structure, np is the number of robots of the structure at the start area, ng as the number of robots in the structure over the void, andnoas the number of robots in the structure on the finish area, the structural stability a structure over the void is satisfied when np=ng=no. Thus, an ng- robot void is able to be crossed when the minimum number of robots in the structure isN=3ng.
B. Simulation Results
Figure 3 shows the steps in the completion of a successful void crossing, when the swarm was able to reach the finish area because its size is satisfying the size of the void and a robot found the target. In this case, eighteen robots have to traverse a 5-robot wide void. Because the minimum robot number required to cross 5-robot void is fifteen robots, the swarm is able to reach the opposite ground. When fifteen robots assembled, the structure was waiting for the sixteenth robot to join so that it could moves forward and maintain its stability when the front robot decoupled.
Figure 4 shows the process by which a swarm crosses a 2-robot wide void, with only three robots available. Unable to detect the opposite ground, the structure moved backward to the start area and the rear robot dispersed from the structure in sequence.
Figure 5 shows the frequency of a specific robot becoming the seed robot and show that all robots have the same oppor- tunity to become the seed. There is also a trend that the robot becoming the seed depends on its initial position, as proved in the case of fixed initial positions. The closer the robots to the void, the higher probability to become the seed robot.
Figure 6 shows the simulation time to cross the void for fixed and random initial positions. As expected, it can be seen that the more robots, the longer the time to cross. This time is determined by the time to find the void, the time to complete the task of locating an object in the finish zone, and actual construction and crossing time.
For the assembly and moving forward time, it is clear that the more robot in the swarm makes the longer time for the swarm to construct the structure and move forward. The success crossing involves the whole process and the structure requires more time to pass the void and make all robots reaching the opposite ground and disassembling to be ready to find the target. The factor of the area dimension does not affect the required time of the process because there is only slightly differences between the recorded time in five case studies. The area dimension was supposed to influence the
(a) Swarm robots at their initial positions. (b) Structure moves forward while recruiting other available robots, while checking for the opposite side of the void.
(c) Fifteen robots are kept in the structure for structural stability, while the leading robots leave the structure.
(d) When there is no robots available to add to the structure, the whole structure moves forward, the leading robot decouples, while maintaining structural stability.
(e) The process continues while the decoupled robots explore the target area.
(f) Task completed as the target has been located and the structure is fully disassembled.
Fig. 3. Eighteen robots crossing a 5-robot wide void with fixed initial position.
(a) Swarm robots at their initial positions. (b) The last free robot joining the structure. (c) The structure moves forward and checks if the opposite zone can be detected
(d) Because the number of robots is not enough to reach the opposite ground, the structure moves backward and the rear robot decouples.
(e) Disassembling of rear robots continues (f) End of simulation
Fig. 4. Example of a failed task, where three robots attempt to cross a void, two robots wide.
Fig. 5. The frequency of individual robot that becomes see robots in a in thirty robot swarm, the higher numbered robots are positioned in the starting area closer to the void. The results are from thirty simulation.
time because free robots are still exploring the area and the distance between them and the structure could be affected by the zone size.
Retreat and disassembling time is only occurred when the swarm failed to cross the void because of unsatisfying number according to the void size. There is a trend that the more the robots in the swarm and the larger the void size, the longer the time required to retreat and disassemble.
For the exploration time at the target zone when the swarm succeed passing the void, it depends only to the random movement of all robots, so there is no trend of increased or decreased graphic. However, the wide area of exploration causes the exploration takes a longer time.
III. ROBOT DESIGN
In the development of a robot suitable for this task, the main challenge is to ensure that the robots can assemble and disassemble autonomously, hence to date effort has been concentrated on this problem. There is main problems the communication requirements and the coupling itself. In the work being considers, it should be noted that the robot has relatively low computing and communication capability, together with a low mass.
A. Previous Work
A wide range of design for robotic assemble-disassemble techniques have been published, of particular note are MIT’s M-Blocks [19] where permanent magnets are embedded in the edges and faces of cubes to passively align and aid the cubes’ movement. This system is effective as it provides a zero power draw with minimal complexity. The Crystal Bot [20] uses hall effect sensors in a similar fashion to M-Block for docking recognition which complements the permanent magnet solution. Further, the system relies on inertial impulse
to detach two units from each other and most swarm robotics could not handle that kind of interaction force. The CATOMS [21] system uses controllable electromagnets to provide a secure attachment on demand. This system is effective in that it can correct relatively high angles of misalignment and on-demand operation means connections are only made when appropriate, as opposed to the M-Blocks. Further, these alignment correcting systems make physical communications a possibility between units in contact, especially where the magnets interface. This is similar to the Molecule [22] and EM-cube [23] systems. However, electromagnets require a constant current draw to maintain the structure, which is an issue for small swarm units. An adaption of electromagnets is to use electro-permanent magnets [24] but these have a high current draw when switching polarity and still present all the noise issues of a permanent magnet solution.
Other systems use mechanical actuators to achieve their interconnections include Roombot [25], ATRON [26] and M- TRAN [27]. Whilst these systems have the potential to be fairly complex due to numerous moving parts at small scales, they tend to also have the benefits of strength and reliability.
The foot-bot is one of three variations of robotic systems developed in the Swarmanoid project [4] and incorporates a scissored actuator that can orient itself around the entire circumference of the circular bot. This actuator slots into the complementary encompassing track on another. This allows for interconnection with numerous configurations of track alignment between bots. Docking recognition is handled via a 2D-force sensor on the gripper. However, these units cannot connect if on any kind of uneven surface next to each other.
The foot-bots uses a top mounted camera to provide a 360o view of local area but still rely on the eye-bots to undertake most detection of the environment for them.
The PolyBot [28] and CONRO [29] systems use a slot and pin to interlock with an SMA actuator that holds the pins in the holes. These pin connections also allow serial communications and docking recognition between docked units. For such applications where some misalignment must be tolerated to handle uneven terrain, these communication systems are not appropriate. Further, SMA actuators have slow deactivation, which is not ideal for controlled and on-demand disassemble of units. A significant current is required to heat the alloy for actuation.
B. Proposed Solution
The design specification can be summarized as follows:
1) Individual robots to be captured and released, on de- mand, as rapidly as possible.
2) The interconnection system must be able to tolerate large degrees of misalignment.
3) The system has minimal power draw throughout use, particularly when in the assembled mode, and during docking.
4) Size of the docking system must be comparable with the mobile platform.
(a) Time for the swarm to locate the void. (b) Time for the swam to start construction, and then retreat because the number of swarm robots presents is less that the number re- quired to cross the void (i.e 15).
(c) Time for the swarm to successfully cross the void.
Fig. 6. The simulation time of both fixed and random initial positions 5-robot void.
Fig. 7. The prototype interconnection mechanism. The robot to the left is coupling to the right hand robot, as the mechanism open, the fingers connect with the loop and pull the robots together, until the end of the worm gear touches the cross member, locking the two robots together.
5) The integrity of the structure must be maintained, with minimum flex in the structure.
6) The robot is capable of discriminating between the environment’s surface and the void.
7) Communication is limited to general broadcast from one robot to the complete swarm, it addition the information content of any message is restricted to a robot’s individ- ual state.
C. Proposed Coupling
The design used in this work is based on a worm-screw based actuator driving geared fingers which open and hooks onto ring on the rear of the leading robot, Figure 7. Once the fingers are opened by the worm-screw the whole assembly will lock the two robots together with very little free movement.
Due to the size of the loop, this approach can accommodate initial misalignment, and does not require any force to be applied to the robot being couple to, hence minimizing the risk of pushing the seed robot over the edge of the void. We have undertaken a FEA analysis of this 3D printed design to confirm that the loop will not fail under the wight of robots during the construction.
D. Proposed Communication
Each robot is provided with six infra red LED receivers and transmitters arranged around the base giving 360o coverage.
To simplify communication requirements, when required, the robot will broadcast a signal at the discrete frequency (i.e 400Hz, 500Hz, etc) depending on the robot’s state and its requirement, typical messages will includevoid locatedandI am the rear robot. This received signal will be processed on using a conventional signal processing approach to determine message. In addition the rear transmitters will provide specific ranging and alignment information during the docking process.
IV. CONCLUSION
In practice a small robot platform cannot cross the void individually, but will be able to complete the task by cooper- atively by constructing a simple line structure, inspired by ant behavior. In this paper, we introduced a strategy, together with the proposed design of a low cost robot, which as a robotic swarm will be able to autonomously cross a void.
The simulations have demonstrated the robustness of the strategy, with various sizes of swarm traversing various sizes of void, or it retreated back to the initial area when the swarm size is not satisfying the stability criteria. Loss a few of individuals robots does not make the overall strategy fail as the same controller in each robot, all robots in the proposed strategy have the same opportunity to be the seed or to find the target so it does not matter if a few number of robots meet the failure to operate. The strategy also shows the flexibility, any number of robots can be added into the swarm and it performs well according to the given main task - finding the target, the swarm is also able to cope with the different size of voids. And finally, the strategy also accommodates the scalability, any size of robotics swarm can perform well. If the number of robots is not enough, robots can retreat to the initial area. If the number of robots is far enough, they can traverse any size of void through a temporary self-assembled dynamic structure.
In further works, this strategy will be implemented in practical using specific platform as robots require strong connection mechanism to assemble each other. By applying a homogeneous robotic swarm, the cost of implementation in
practical can be reduced. In the case of large void crossing, a similar strategy is being developed by implementing a complex structure to reduce the deflection effect of the structure so that robots can safely reach the opposite area.
ACKNOWLEDGMENT
The authors acknowledge the for funding of the scholarship for Niken Syafitri from the Directorate General of Resources for Science, Technology and Higher Education Ministry of Research, Technology and Higher Education of Indonesia.
REFERENCES
[1] D. M. Bonabeau E. and T. G., Swarm Intelligence: From Natural to Artificial Systems. Oxford University Press, New York, 1999.
[2] M. Brambilla, E. Ferrante, and M. Birattari, “Swarm robotics: A review from the swarm engineering perspective.”Swarm Intelligence, vol. 7, no. 1, pp. 1–41, 2012. [Online]. Available: http://link.springer.com/10.
1007/s11721-012-0075-2
[3] G. Beni, “From Swarm Intelligence to Swarm Robotics,” in Swarm Robotics WS 2004, E. Sahin and W. Spears, Eds. Springer-Verlag Berlin Heidelberg, 2005, vol. 3342, pp. 1–9. [Online]. Available:
http://link.springer.com/content/pdf/10.1007/978-3-540-30552-1 1.pdf [4] M. Doriigo, D. Floreano, L. M. Gambardella, F. Mondada, S. Nolfi,
T. Baaboura, M. Birattar, M. Bonani, M. Brambilla, A. Brutschy, D. Burnier, A. Campo, A. L. Christensen, A. Decugni`ere, G. D.
Caro, F. Ducatelle, E. Ferrante, A. F¨orster, J. Guzzi, V. Longchamp, S. Magnenat, J. M. Gonzales, N. Mathews, M. M. de Oca, R. O’Grady, C. Pinciroli, G. Pini, P. R´etornaz, J. Roberts, V. Sperati, T. Stirling, A. Stranieri, T. St¨utzle, V. Trianni, E. Tuci, A. E. Turgut, and F. Vaussard,
“Swarmanoid: A Novel Concept for the Study of Heterogeneous Robotic Swarms,”IEEE Robotics & Automation, no. December, pp. 60–71, 2013.
[5] S. Kernbach, F. Schlachter, R. Humza, J. Liedke, S. Popesku, S. Russo, T. Ranzani, L. Manfredi, C. Stefanini, R. Matthias, C. Schwarzer, B. Girault, P. Alschbach, E. Meister, and O. Scholz, “Heterogeneity for Increasing Performance and Reliability of Self-Reconfigurable Multi- Robot Organisms,” IROS11, Workshop on Reconfigurable Modular Robotics, 2011. [Online]. Available: http://arxiv.org/abs/1109.2288 [6] T. Kuyucu, I. Tanev, and K. Shimohara, “Hormone-Inspired Behaviour
Switching for the Control of Collective Robotic Organisms,”Robotics, vol. 2, no. 3, pp. 165–184, 2013. [Online]. Available: http:
//www.mdpi.com/2218-6581/2/3/165/
[7] P. Levi, E. Meister, and F. Schlachter, “Reconfigurable swarm robots produce self-assembling and self-repairing organisms,” Robotics and Autonomous Systems, vol. 62, no. 10, pp. 1371–1376, 2014. [Online].
Available: http://dx.doi.org/10.1016/j.robot.2014.07.001
[8] H. Li, H. Wei, J. Xiao, and T. Wang, “Co-evolution framework of swarm self-assembly robots,”Neurocomputing, vol. 148, no. 2015, pp.
112–121, 2015. [Online]. Available: http://dx.doi.org/10.1016/j.neucom.
2012.10.047
[9] V. Von´asek, M. Saska, L. Winkler, and L. Peuˇcil, “High-level motion planning for CPG-driven modular robots,” Robotics and Autonomous Systems, vol. 68, pp. 116–128, 2015. [Online]. Available:
http://dx.doi.org/10.1016/j.robot.2015.01.006
[10] R. Bruni, A. Corradini, F. Gadducci, A. Lluch Lafuente, and A. Vandin,
“Modelling and analyzing adaptive self-assembly strategies with Maude,”Science of Computer Programming, vol. 99, pp. 75–94, 2015.
[11] R. O’Grady, R. Groß, A. L. Christensen, and M. Dorigo, “Self-assembly strategies in a group of autonomous mobile robots,”Autonomous Robots, vol. 28, no. 4, pp. 439–455, 2010.
[12] R. O’Grady, A. L. Christensen, and M. Dorigo, “SWARMORPH:
Multirobot Morphogenesis Using Directional Self-Assembly,” IEEE Transactions on Robotics, vol. 25, no. 3, pp. 738–743, 2009.
[13] V. Trianni, E. Tuci, C. Ampatzis, and M. Dorigo, “Evolutionary Swarm Robotics: a theoretical and methodological itinerary from individual neuro-controllers to collective behaviours,”
in The Horizons of Evolutionary Robotics, P. Vargas, E. Paolo, I. Harvey, and P. Husbands, Eds., MIT Press, 2014, no. January 2011, pp. 153–178. [Online]. Available:
http://mitpress.mit.edu/books/horizons-evolutionary-robotics%5Cnhttp:
//laral.istc.cnr.it/Pubblicazioni/English/Journals/2010trianni-etal-her.pdf [14] L. Cucu, M. Rubenstein, and R. Nagpal, “Towards self-assembled
structures with mobile climbing robots,” in Proceedings of the 2015 IEEE International Conference on Robotics and Automation, 2015, pp.
1955–1961.
[15] E. Carpanzano, D. Dallefiate, and F. Jatta, “A Modular Framework for the Development of Self-reconfiguring Manufacturing Control Systems,”
in Proceedings of the 2002 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems, 2002.
[16] J. M. Montanier and P. C. Haddow, “Adaptive self-assembly in swarm robotics through environmental bias,” in Proceedings of the IEEE Symposium Series on Computational Intelligence - IEEE ICES: 2014, 2014, pp. 187–194.
[17] K. H. Yoo, Y. Lee, H. J. Cho, B. H. Yoo, and D. H. Kim, “Swarm Robotics: Self Assembly, Physical Configuration, and Its Control,” in SICE-ICASE International Joint Conference 2006, 2006, pp. 4276–4279.
[18] R. Crowder and K.-P. Zauner, “A project-based biologically-inspired robotics module,”IEEE Transactions on Education, vol. 56, no. 1, 2013.
[19] J. W. Romanishin, K. Gilpin, S. Claici, and D. Rus, “3D M-Blocks:
Self-reconfiguring robots capable of locomotion via pivoting in three dimensions,” in2015 IEEE International Conference on Robotics and Automation (ICRA), 2015, pp. 1925–1932. [Online]. Available: http:
//ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=7139450 [20] D. Rus and M. Vona, “A basis for self-repair robots using
self-reconfiguring crystal modules,” Intelligent Autonomous Systems, pp. 2194–2202, 2000. [Online]. Available: http://citeseerx.ist.psu.edu/
viewdoc/download?doi=10.1.1.28.8956&rep=rep1&type=pdf
[21] S. C. Goldstein, “Programmable Matter,”Computer, no. June, pp. 99–
101, 2005.
[22] C. McGray and D. Rus, “Self-reconfigurable molecule robots as 3D metamorphic robots,” in Proceedings. 1998 IEEE/RSJ International Conference on Intelligent Robots and Systems. Innovations in Theory, Practice and Applications (Cat. No.98CH36190), vol. 2, 1998, pp. 0–5.
[23] B. K. An, “EM-Cube: Cube-shaped, self-reconfigurable robots sliding on structure surfaces,” inProceedings - IEEE International Conference on Robotics and Automation, 2008, pp. 3149–3155.
[24] A. D. Marchese, H. Asada, and D. Rus, “Controlling the locomotion of a separated inner robot from an outer robot using electropermanent magnets,” inProceedings - IEEE International Conference on Robotics and Automation, 2012, pp. 3763–3770.
[25] M. Vespignani, E. Senft, S. Bonardi, R. Moeckel, and A. J. Ijspeert,
“An experimental study on the role of compliant elements on the locomotion of the self-reconfigurable modular robots Roombots,”
in 26th IEEE/RSJ International Conference on Intelligent Robots and Systems:, 2013, pp. 4308–4313. [Online]. Available: http:
//dx.doi.org/10.1109/IROS.2013.6696974
[26] M. Jorgensen, E. Ostergaard, and H. Lund, “Modular ATRON: mod- ules for a self-reconfigurable robot,” in 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IEEE Cat.
No.04CH37566), 2004, pp. 2068–2073.
[27] S. Murata, E. Yoshida, A. Kamimura, H. Kurokawa, K. Tomita, and S. Kokaji, “M-TRAN : Self-Reconfigurable Modular,” IEEE/ASME Transactions on Mechatronics., vol. 7, no. 4, pp. 431–441, 2002.
[28] M. Yim, Y. Zhang, K. Roufas, D. Duff, and C. Eldershaw, “Con- necting and disconnecting for chain self-reconfiguration with polybot,”
IEEE/ASME Transactions on Mechatronics, vol. 7, no. 4, pp. 442–451, 2002.
[29] A. Castano and P. Will, “Mechanical design of a module for recon- figurable robots,” inProceedings of the 2000 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2000), vol. 3, 2000, pp. 2203–2209.
