Helping Hands: Swarm Robotics
The Intersection of Swarm Robotics and Physical Sciences
Thomas Jefferson High School for Science and Technology
This article was originally included in the 2018 print publication of Teknos Science Journal.
I groaned in frustration as I realized my simulation hit another dead end. Looking back at my spreadsheet, I prepared to run yet another simulation. After modifying the parameters, I restarted the simulation software. A model of a forest appeared on the screen with two bright red targets positioned randomly among the clusters of trees. Just seconds later, three Parrot AR.Drones drones spawned onto the screen. Keeping a close eye on the debugging information, I pressed a couple keys and let the simulation start. The drones began moving around the field, carefully scanning and recording everything in their paths to locate both of the targets as quickly as possible.
This intelligent coordination of a multi-robot system is known as swarm robotics. With the increase in technological advances in computer science and robotics, scientists have sought to simultaneously control swarms of robots to achieve certain tasks. Typically, a task can be completed by one robot or a group of robots. Performing the task using multiple robots allows for split workloads, thus giving swarm robots more functionality and efficiency when compared to the performance of a single robot. Swarm robotics has many applications in area exploration, specifically in situations including search and rescue applications and military reconnaissance. Additionally, swarm robotics has applications in the entertainment industry. For example, Intel organizes annual choreographed light shows with hundreds of drones. Whatever the applications may be, it is becoming increasingly evident that swarm robotics will be vital in the future.
An essential part of any successful multi-robot system is the algorithm used to control the robots in tandem. Algorithms compiled by Bayinder (2016) suggest the majority of swarm robotics algorithms use one of the following design methods to control robot movement: application of virtual force (physicomimetics), nature-inspired method, and artificial evolution. Each design method has its own benefits and drawbacks, leading scientists to often prefer one over the others for their own specific purposes. For example, artificial evolution algorithms are sometimes preferred due to their high flexibility in implementation and programming, making coding much easier. In fact, Bakhshipour, Ghadi, and Namdari (2017) conducted research to create a novel artificial intelligence-inspired algorithm for search and rescue applications to avoid the drawbacks of using a typical nature-inspired approach to swarm robotics due to the lack of simultaneous complex interactions in nature.
Ultimately, application of virtual forces and nature-inspired methods are the most common design methods for controlling swarm movements. The physicomimetic approach simplifies robot interactions to follow laws analogous to real physical science. Research from Reynaud and Guerin-Lassous (2016) highlights the possible applications this approach. In their paper, they outlined the implementation of a force-controlled multi-drone system for pest management, using Virtual Force Protocol. With the Virtual Force Protocol, in addition to the traditional forces of gravity, thrust, and drag, the robots are subjected to an imaginary force from their neighboring robots that either attract or repel the robots. Unfortunately, using the laws of physics to model swarms of robots can often be limiting due to the laws’ inflexible nature, as opposed to the flexibility of animal behaviors refined by thousands of years of natural selection.
A nature-inspired approach (NIA) is often preferred because of its diversity and robustness resulting from natural selection. A nature-inspired approach simplifies robot interactions to follow swarm interactions found in nature, such as the foraging behavior of honey bees, the echolocation characteristics of microbats, or the pheromone trail laying behavior in ant colonies (Bayinder, 2016). Research conducted by Yang, Ding, Jin, and Hao (2015) indicates the effectiveness of a nature-inspired swarm algorithm approach using bacterial chemotaxis behavior. The scientists were able to develop their novel algorithm and run simulations by comparing the effectiveness with traditional algorithms. Ultimately, they discovered that their algorithm had a very low overhead cost, meaning the algorithm did not require much computation power, and that the algorithm had a low chance of forcing robots into local optimums, a problem I have encountered in my own research.
With my research at the Automation and Robotics Lab at TJHSST, I hope to create an effective hybrid algorithm that utilizes a physicomimetic approach, with virtual forces, and nature-inspired approach, virtual insect pheromone trails, for robot movements in a search and rescue application. A hybrid algorithm attempts to counterbalance the disadvantages with a physicomimetic and nature-inspired approach. The primary obstacle in the creation of my hybrid algorithm is determining how to limit the number of local minima, which occur when robots get stuck due to some interaction laws opposing each other equally. For example, a robot could get stuck between two other robots as the outer robots push back on the inner robot with equal “forces.”
Fortunately, the problem with local optimums can be avoided. Research done by Sharma, Shukla, and Tiwar (2016) underscores the effectiveness of a similar hybrid algorithm that also avoids this issue. They designed a hybrid algorithm using a Clustering Based Distribution Factor (CBDF), assigning each swarm robot a separate partition of the map and a nature-inspired algorithm, using movements inspired by bacteria foraging and bat flight patterns. The researchers found the hybrid algorithm to be extremely effective with the CBDF able to avoid the obstacle of getting stuck in local minimas. In order to avoid the issue of getting stuck in local minimas, I contacted Dr. Sharma. He revealed several critical components of the CBDF that allowed his team to avoid getting stuck in local minimas. Specifically, he recommended using “Zig-Zag Search Effect (ZSE), in which the direction to the final cluster target changes frequently and dynamically. [ZSE] offers an escape path from a local minima by applying random velocity vectors” (personal communication, Jan. 21, 2018). Sharma’s information will help me eliminate possible drawbacks in algorithm through bypassing local minimas. A hybrid swarm robotics algorithm that compensates for the respective drawbacks with each individual design method could have valuable applications in military reconnaissance and search and rescue, with the potential to save human lives.
Bakhshipour, M., Ghadi, M. J., & Namdari, F. (2017). Swarm robotics search & rescue: A novel artificial intelligence-inspired optimization approach. Applied Soft Computing, 57, 708-726. https://doi.org/10.1016/j.asoc.2017.02.028
Bayindir, L. (2016). A review of swarm robotics tasks. Neurocomputing, 172, 292-321. https://doi.org/10.1016/j.neucom.2015.05.116
Reynaud, L., & Guerin-Lassous, I. (2016). Design of a force-based controlled mobility on aerial vehicles for pest management. Ad Hoc Networks, 53, 41-52. https://doi.org/10.1016/j.adhoc.2016.09.005
Sharma, S., Shukla, A., & Tiwari, R. (2016). Multi robot area exploration using nature inspired algorithm. Biologically Inspired Cognitive Architectures, 18, 80-94. https://doi.org/10.1016/j.bica.2016.09.003
Yang, B., Ding, Y., Jin, Y., & Hao, K. (2015). Self-organized swarm robot for target search and trapping inspired by bacterial chemotaxis. Robotics and Autonomous Systems, 72, 83-92. https://doi.org/10.1016/j.robot.2015.05.001