Lecture
Swarm robotics is an approach to coordinating multiple robots as a system consisting of a large number of mostly simple physical robots. It is assumed that the desired collective behavior emerges from the interactions between robots and the interactions of robots with their environment. This approach emerged in the field of artificial swarm intelligence, as well as in biological studies of insects, ants, and other areas in nature where swarm behavior occurs.
The study of swarm robotics focuses on the design of robots, their physical bodies, and their controlling behavior. It was inspired by, but is not limited to, the emergent behavior observed in social insects, known as swarm intelligence. Relatively simple individual rules can give rise to a large set of complex swarm behaviors. A key component is communication between group members, which builds a system of continuous feedback. Swarm behavior involves the constant change of individuals in cooperation with others, as well as the behavior of the whole group.
Unlike distributed robotic systems in general, swarm robotics emphasizes a large number of robots and promotes scalability, for example by using only local communication. This local communication can be achieved, for example, using wireless transmission systems such as radio frequency or infrared.
The study of swarm robotics — is the study of robot design, their appearance, and behavior control. Its emergence is related to (but not limited to) the systemic effect of behavior observed in social insects and called swarm intelligence. Relatively simple rules of individual behavior can create complex, organized behavior of the whole swarm. The key point is the interaction between group members, which creates a system of continuous feedback. Swarm behavior includes the constant change of participants interacting with each other, as well as the behavior of the whole group as a whole.
Unlike simply distributed robotic systems, swarm robotics emphasizes a large number of robots, and also assumes scalability, for example using only local communication. This local communication can be implemented, for example, on the basis of wireless data transmission systems in the radio-frequency or infrared range.
An important tool for the systematic study of group behavior is video tracking, although other tracking methods also exist. Recently[when?] the Bristol Robotics Laboratory developed an ultrasonic swarm tracking system for research purposes. Further research is needed to find methods that will provide a reliable prediction of group behavior when only the properties of individual members are given.
Miniaturization and cost are key factors in swarm robotics. These are constraints when creating large groups of robots; therefore the simplicity of an individual team member should be emphasized. This should motivate an intelligent approach to the swarm in order to achieve meaningful behavior at the swarm level, rather than at the individual level.
Much research has been aimed at achieving simplicity at the level of individual robots. The ability to use real hardware for Swarm Robotics research, rather than simulation, allows researchers to discover and solve far more problems, as well as expand the scope of Swarm Research. Thus, developing simple robots for swarm exploration research is a very important aspect of this field. Goals include reducing the cost of individual robots to ensure scalability, making each swarm member less resource-demanding and more energy-efficient.
Compared to individual robots, a swarm can usually decompose a given mission into subtasks; a swarm is more resilient to partial swarm failure and more flexible with regard to different tasks.
One such swarm system is the LIBOT robotic system, which uses an inexpensive robot designed for outdoor robotics. The robots are also equipped for indoor use over Wi-Fi, since GPS sensors provide poor connectivity indoors. Another such attempt is the microrobot (Colias), built at the Computer Intelligence Laboratory of the University of Lincoln, UK. This microrobot is built on a 4-centimeter round chassis and represents an inexpensive open platform for use in various Swarm Robotics applications.
Swarm robotics has many potential applications. These include tasks requiring miniaturization (nanorobotics, microbotics), such as distributed sensing tasks in micromachines or the human body. One of the most promising applications of swarm robotics is disaster rescue operations. Swarms of robots of various sizes can be sent to places rescuers cannot safely reach, to detect the presence of life using infrared sensors. On the other hand, swarm robotics may be suited to tasks requiring cheap designs, for example mining and agricultural foraging tasks.
More controversially, swarms of military robots could form an autonomous army. The US Navy has tested a swarm of autonomous boats that can independently maneuver and carry out offensive actions. The boats are unmanned and can be fitted with any equipment for deterring and destroying enemy vessels.
During the civil war in Syria, Russian forces in the region reported attacks on their main air base in the country by a swarm of fixed-wing aircraft loaded with explosives. [10]
Most efforts have focused on relatively small groups of machines. However, in 2014 Harvard demonstrated a swarm consisting of 1024 individual robots, which became the largest to date. [11]
Another large set of applications can be addressed using swarms of micro air vehicles, which are also being widely researched at present. Compared to pioneering studies of flocks of flying robots using precise motion-capture systems in laboratory conditions [12], modern systems such as Shooting Star can control groups of hundreds of micro vehicles outdoors [13] using GNSS systems (such as GPS), or even stabilize them using onboard localization systems [14] where GPS is unavailable. [15] [16] Swarms of micro air vehicles have already been tested in tasks of autonomous surveillance [17], plume tracking [18], and reconnaissance in tight formation. [19] Numerous works have been carried out on cooperative groups of unmanned ground and aerial vehicles with target applications of cooperative environmental monitoring, [20] simultaneous localization and mapping, [21] convoy protection, [22] and locating and tracking moving targets. [23]
In addition, progress has been made in applying autonomous swarms in manufacturing, known as swarm 3D printing. This is particularly useful for manufacturing large structures and components, where traditional 3D printing cannot be used due to equipment size limitations. Miniaturization and mass mobilization allow the manufacturing system to achieve scale invariance, without being limited by the effective assembly volume. Being at an early stage of development, swarm 3D printing is currently being commercialized by startups. Using a metal additive manufacturing process based on rapid induction printing, Rosotics [24] was the first company to demonstrate swarm 3D printing using a metal payload, and the only one to achieve metal 3D printing from an onboard platform. [25]
The foundation of the software code of the future, i.e. diffuse applied code, is based on three main principles:
For example, the BEECLUST algorithm was made by us based on information about the behavior of young bees in complex temperature conditions. This collective behavior allows bees to always find the spot with the temperature most suitable for them (global optimum) among all warm spots (local optima). The basic principles of interaction were extracted from videos of our experiments with real bees, which we conducted in our laboratory. These principles were then expressed as a simple computer algorithm, which was embedded in a group of autonomous robots, allowing them to jointly find such optima in their environment. BEECLUST — is the simplest of the currently existing swarm algorithms, and yet it is quite effective.
Communication must be established between the robots. We work on developing these components in collaboration with our partners on the technology side. Fundamentally, we are a biological laboratory and work together with colleagues from engineering fields — mechatronics, electronics, and sensor technology.
For the SYMBRION and REPLICATOR projects, several dozen cubic robots were built (approximately 10×10×10 cm), each of which was an autonomous robot that moved using tracks or screws. All elements had joints that allowed them to bend, and four docking ports that allowed them to physically connect to other robots, as well as disconnect. Thanks to this, the robots can form a group of cells that connect to form more complex "organisms" of various kinds and shapes.
To control this process of "embryogenesis" we developed a program called "Virtual Embryogenesis" (VE for short), to which my colleague Dr. Ronald Thenius contributed a great deal. This program is based on knowledge of animal embryogenesis. In VE, the robots embody a model of biological embryogenesis, in which gene activation entails the production of substances (morphogens) that spread and create concentration gradients in the body of the embryo. Subsequently, the local concentration of such morphogens determines the activation or blocking of the expression of other genes. This system is a very complex cascade of interactions between genes and morphogens, which can, in essence, be regarded as a true "program" according to which embryogenesis proceeds. In VE, the robots act according to such a model of processes, "turning on" and "turning off" virtual genes depending on the local concentration of morphogens, which is in turn affected by gene activation. These levels of virtual morphogen concentration also depend on a diffusion-like process of communication between neighboring robots. Ultimately, the concentrations of the virtual morphogen affect the docking ports, and in this way the multi-modular robotic organism can grow and reorganize.
To allow such a robotic "multicellular" organism to exhibit behavioral patterns such as walking, we developed another program, which we called the "Artificial Homeostatic Hormone System" (AHHS), also known as the "hormone controller". This controller is similar in its mechanism of action to the virtual embryogenesis described above. Each element secretes (releases) substances that act as hormones into its surroundings. These substances are locally added to and removed from the system according to rules that can be implemented based on the local concentration of hormones. These rules are analogous to the element’s "genome". In contrast to "Virtual Embryogenesis", which is used to "grow" the organism by coordinating the docking of individual robots, AHHS is used to allow the organism to move. Hormone concentrations are also influenced by information from sensors, and they, in turn, affect the elements’ locomotion mechanisms. Thus, this represents an information-processing system that drives the whole robotic organism.
One advantage of this system is that the resulting hormone concentration gradients can also divide the organism into parts in which different programs can run. For example, the legs can run a program different from the one governing the spine, regardless of how many legs the organism has. And, based on information coming from various sensors, the organism can switch between programs. For example, if the battery is running low, a "hungry" organism may move differently than a "full" organism. Thus, this program, built on the model of the animal hormonal regulation system, takes into account the shape of the body, reacts to environmental conditions via sensors, and can control behavior or the decision-making mechanism for how to act. This system was developed by me in collaboration with my colleagues — Dr. Heiko Hamann, Dr. Jürgen Stradner, and Dr. Payam Zahadat.
Using the Evo-Devo approach (evolution of development), which is a combination of developmental biology and evolutionary theory. According to this approach, embryogenesis and evolution are interconnected processes, since evolution is based on genetic adaptations that entail adaptation of the developing organism, and how well-adapted the organism is to conditions affects its evolutionary fate. In our methods, such as the VE and AHHS described above, all programs/rules are embedded in a data structure called the "genome", and by applying evolutionary computation techniques we adapt these genomes depending on how well the organism performs. We also make use of ecological knowledge when developing such robot systems, since we have always had many artificial organisms "living" in the same habitat, which could give rise to cooperation and competition. We developed a series of experiments in which we placed organisms in conditions where they had to either compete or cooperate with other organisms. In this way, the evolution of cooperation occurred indirectly, since only those that successfully cooperated or competed reached the final goal.
The SubCULTron project will be the first case of applying such a large number (150+) of coordinated autonomous robots in the real world to monitor the Venice Lagoon. The current state of affairs is such that if a group of robots is large, it operates only within the laboratory, while only a very small number of robots operate in the field. However, even such trials outside the laboratory are short-lived and partly controlled by researchers. In our subCULTron project, we went much further: the robots will operate in the lagoon for a long time — several days or even weeks — and will be almost 100% autonomous.
In other projects, ASSISIbf and FloraRobotica, swarm algorithms are used in combination with living organisms — real bees, ants, or plants. In the future this could help in animal husbandry, animal monitoring, and agriculture. For example, in ASSISIbf we combine robots with live bees and tune behavioral feedback mechanisms. This means that the bees influence the robots, and the robots, in turn, influence the bees, which ultimately leads to the merging of two "societies" — robots and bees — into one. In addition to this, we use evolutionary computation and machine learning to adapt the robots to bee behavior.
At the moment there are two main research goals here:
1) to evolve a program that would allow a group of robots to roughly estimate the local density of bees around them, that is, this is a task in the evolution of collective perception;
2) to obtain, using the same method, a program that would make the bees perform a required task, such as gathering in a specific place.
This is a task in the evolution of collective action. The ability to obtain information from/about a population of animals, as well as the ability to control them (without coercion), can be very useful in keeping and managing animals, even parasitic ones. Among other things, this can be a more humane way of treating animals, since we are thereby not forcing them to do something, but "persuading" them to do it — for example, gathering them in a certain place using stimuli familiar to them. Our robots pose as such animals themselves and thus gain a voice in the animal community.
Evolutionary computation has one major problem, namely the complexity of the behavior it produces. For example, it is fairly easy to evolve programs that avoid collisions or find a target, but with higher-level behavior everything is much more difficult. Evolution usually favors the simplest ("cheapest") solution that fits in one way or another.
Swarm robotics is built on creating complex things from small simple elements. And evolutionary swarm robotics is a promising direction, since it combines a mutually beneficial combination of swarm robotics and evolutionary computation. However, this field is still little studied, since it is difficult to set up the necessary structures. The most interesting subtype of evolutionary swarm robotics is the use of continuous embedded evolution in real robots. This means that the evolutionary algorithm controls all the robots simultaneously in real time. There are quite a few technical nuances here, since this requires good decentralized communication and long operating time.
As mentioned earlier, self-reconfiguring, self-programming, self-adapting groups of robots will be excellent tools for exploring uncharted territories with potentially harsh conditions. The most obvious examples are, of course, the ocean depths and other planets. In the near future, applications will be in the field of environmental monitoring, or in military affairs, or for the entertainment industry. So there is also a "toy factor" here, so I expect that toys based on swarm robotics will appear soon and that people will start playing with them and thinking about what else can be done with them, so there is also a certain outreach/educational component here.
Drone displays typically use several illuminated drones at night for artistic display or advertising.
The main plot of Disney’s "Big Hero 6" involved the use of swarms of microbots to form structures.
Swarm robotics is used in the Tamil film Enthiran and its sequel 2.0.
Comments