Artificial robot organism

Artificial Robot Organisms - is a new research field within the domain of swarm, evolutionary and reconfigurable robotics. Some first works go to early 90s in the field of cellular robotics, lately it is performed by several Japan, American and European teams of researchers. Different aspects of this work are considered in several research projects , , , lately European Commission supported new research initiatives related to a new generation of artificial robot organisms , .
New Paradigm in Collective Robotic Systems
Collective intelligence is often associated with macroscopic capabilities of coordination among robots, collective decision making, labor division and tasks allocation in the group . The main idea behind this is that robots are achieving better performance when working collectively and so are capable of performing such activities which are not possible for individual robots. The background of collective intelligence is related to the capability of swarm agents to interact jointly in one medium. There are three different cases of such interactions:
1. In the first case agents communicate through a digital channel, capable for semantic messages exchange. Due to information exchange, agents build different types of common knowledge . This common knowledge in fact underlies collective intelligence.
2. The second case appears when macroscopic capabilities are defined by environmental feedback. The system builds a closed macroscopic feedback-loop, which works in a collective way as a distributed control mechanisms. In this case there is no need of complex communication, agents interact only by kinetic means. This case if interaction is often denoted as a spatial reasoning, or spatial computing.
3. The third case of interactions we encounter in nature, when some bacteria and fungi (e.g. dictyostelium discoideum) can aggregate into a multi-cellular organism when this provides better chances of survival . In this way, they interact not only through information exchange or spatial interactions, they build the closest physical, chemical and mechanical interconnections, through the agents still remain independent from each other . The first two cases of interactions are objects of extensive research in many domains: robotics, multi-agents systems, bio-inspired and adaptive community and so on. However, the practical research in the last case represents essential technological difficulties and therefore is not investigated enough. Despite the similarities between a robot swarm and multi-robot organism, such as a large number of robots, focus on collective/emergent behavior, a transition between them is a quite difficult step due to mechanical, electrical and, primarily, conceptual issues. Now, the research around the third case of interactions is concentrated on three important questions:
1. Reconfigurability, adaptability and learn-ability of symbiotic systems. These issues include flexible and multifunctional sensors and actuators, distributed computation, scalability, modelling, control and other issues, which are closely related to the reconfigurable robotics.
2. Evolve-ability of symbiotic systems, which includes principles and aspects of long- and short-term artificial evolution and adaptivity as well as exploring and analogies to biological systems.
3. Embodiment of evolutionary systems for different environments and medias as well as investigation of information properties of such systems.
In this way, a transition from robot swarm to multi-robot organisms represent one of possible next steps in a further research within the collective robotic community.
Differences to Classical Robotics
In robotics, a robot typically consists of a computational core (one or many microcontrollers, microprocessors or computers connected via different networks), central or distributed sensor system and, finally, actuators. Actuators are spatially distributed so that to achieve a possibly effective actuation (and locomotion). Such a robotic system can be controlled in two different ways:
* in classical robotics, there is one central system, which processes all available sensor data and gives commands to actuators. In many different systems we can observe several variations of this principle, like a cooperative actuation of industrial manipulators or modern semi-autonomous AGV.

* in collective robotics, all robots first “negotiate” via communication network in order to process the data collectively and to reach agreement about common activities. The capabilities and the way of how the robots reach coordination in the group can be also different: networked robotics uses global communication to achieve a desired coordination; whereas swarm robots are typically more limited in capabilities and uses local interactions for creating their collective behaviour.

Adaptation of these robot systems to new environments usually consists in a change of parameters, reconfiguration of tools, change of a cooperation pattern between robots, using of software-based learning mechanisms. Capabilities and functionality, especially actuation, of these robotic systems are basically defined by corresponding hardware design and can hardly be changed during the run-time (i.e. without a human assistance). However in real situation, we often observe a need to make a flexible sensing, computational and actuating functionality, which can be changed in a run-time situation in humanless environments.

Appearance of artificial robot organisms is one of possible answers to this challenge. Robot organism operates first as a collective (or swarm-like) system, consisting of many independent, fully autonomous robot-elements (see image right). The elementary robots are able to move, to communicate, to make decisions and so on. However, these robots are able to dock to each other and to perform the following activities in the aggregated state (see image left):
* share energetic recourses and perform a common energy management. It includes energy re-distribution, storage and re-allocation within the whole organisms;
* communicate through high-speed internal wired network to exchange sensor data, use distribute memory mechanism, use computational resources in parallel;
* perform a joint actuation, which allows increasing common degrees of freedom for the organisms. For example, combining a few 1DoF elements, the whole system is able to achieve the required degrees of freedom.

In this way we observe a new system, whose control principles differ from classical and collective robotics:
* robots are capable for autonomous aggregation and disaggregation; elementary robots in the disaggregate state possess individual locomotion;
* many aggregated robots in the organism operate as one common entity. For efficient behavior, organisms need centralized or semi-centralized control system. However, elementary-robots within the organism still remain independent, their internal homeostasis is still closed and separated from each other;
* new, organism-relevant mechanisms appear, for example, artificial immune system to monitor and defend the organism from “infections”, artificial genetic system, which manages and stores organism-relevant information, organism’s energetic homeostasis, massive sensor data fusion and others.
* organism-relevant and individual-relevant control mechanisms co-work. Moreover, we say that elementary robots co-exists in a symbiotic way in the organism, therefore the Artificial Robot Organisms calls sometimes Symbiotic Robotics.
* due to a cellular construction, robot organisms are enormously flexible and reliable. This flexibility leads to the challenge of a controllable self-development of a robot and phenomenon of artificial evolution of these organisms.
Towards Evolve-ability of the Robot Organism
Evolve-ability is one of the most important issues in the research around artificial organisms. From the conceptual point of view, the evolve-ability can be achieved in two complementary ways, which can be denoted as bio-inspired (or bio-mimicking) and engendering-based approaches.
Bio-inspired/bio-mimicking approach
Any bio-inspired approach is based on analogies to living organisms. Bio-inspired control algorithms use neither any global point of information nor any form of complex knowledge. These algorithms are stable to a wide range of environmental conditions and are extremely robust. Therefore, the bio-inspired strategies are going to draw advantage from the well-known robustness/simplicity as well as from the plasticity/adaptability derived from natural systems. The goal is to create stable, robust and adaptable robotic organisms. Here we will investigate a variety of concepts, such as:
* Genome: All robotic organisms will carry one or several Genomes. A Genome is a collection of genes, which carry information about controller structure and controller dynamics. A gene can be a simple part of a blueprint, which ”depicts” a part of the final controller. But a gene can also work as a rule, which is used to ”construct” parts of the final controller. In the latter case, there can be interferences between different genes, thus competition or cooperation can arise also on the genetic level. A self-organized process can be established which will be able to create a flexible, but robust controller structure.
* Controller: Several controller types can be applied to a behavioral control, ranging from rules-based controllers, to Evolvable Artificial Neural Networks (EANN) and Artificial Immune Networks (AIS), to hormone-based controllers and to even hand-coded controllers that execute hand-optimized (modular) parts of the whole organism’s behavioural repertoire.
* Sexuality/Reproduction: Dynamics of artificial evolution can be enhanced by implementing virtual-reproduction of robots. A separate process will allow removing controllers from the least fit robots and to re-initialize them with mixtures (interbreeds) of the controllers of more fit robots, this "sexual" reproduction provides advantages in several scenarios.
* Embryology: To allow well-ordered controllers to emerge from the information stored in the Genome, embryological processes can be mimicked, driven by a virtual hormone system.
Engineering-based approach
The engineering-based approach is complementary to the bio-inspired one and focuses in such issues as learning, distributed decision making, navigation and so on. Generally, consortium focuses on three following approaches (these approaches are closely connected so that finally it will be a kind of hybrid framework):
* On-line learning. On-line learning is based on the behavior level and uses automatically generated feedback. The feedback comes from internal, external and virtual sensors. Some direct feedback can be sensed through vision-based subsystem, by using FRID-based identification or localization technologies, by using smart laser scanner, sound, light, humidity, temperature, internal energy sensor and other sensors. It is intended to use middleware and sensor-fusion approach to generate complex non-direct feedbacks through virtual sensors. Since off-line mechanisms can hardly be applied to real robots, the challenge of the proposed approach is to perform non-supervised learning without any off-line mechanisms (or at least with a minimum of them). This can be achieved by combining evolving computation with re-wards/feedback/fitness calculated on-line. Therefore the whole approach can be named ”on-line learning”.
* Evolutionary computation. High computational power of the system allows running on-line and on-board such well-known approaches as genetic programming (GP) (e.g.), Genetic Algorithms (GA) (e.g. ). To avoid the problems posed by a huge search space, it is intended to integrate limitations, originating from hardware platform. Another set of problems are the fitness functions required for these algorithms. These fitness functions are very difficult to calculate based only on local sensor data. Moreover these functions are evaluated extremely delayed because the organism mostly assess their fitness after accomplishing the task.
* Approaches from the domain of Distributed Artificial Intelligence (DAI). On-line learning as well as GA/GP include diverse aspects of DAI such as a distributed knowledge management, semantic information processing, navigation and actuation in the environment, planning, sensor fusion and others. Development and implementation of these approaches is an important step towards evolve-ability of the robot organisms.
Challenges
* Technological challenges, for example of making a possibly simple, but reliable mechanical docking system, capable of electrical connection for high-speed data transfer, a few watt of power transmission and enough mechanical torque of lifting several other modules
* Truly symbiotic multi-cellular construction of real-world artificial organisms. Elementary robots equivalent to single cells will build artificial-life-forms with a central nervous system, common energy resources and homoeostasis at the level of the whole organism.
* Principles and aspects of long- and short-term artificial evolution together with evolve-ability and adaptivity for real multi-agent systems with symbiotic principles of self-organisation and emergence.
* Fundamental questions such as “how many cells with different DNA can form an organism with one common DNA?”, “how does specialization of cells within the organism appear?”, “how can 'cancer cells' appear within the organism?” - i.e. questions which are open and highly relevant in both scientific and human contexts.

Main Menu

See Also

Main Menu