Intelligent Robotics Group: Biologically Inspired Robotics

Lifelong Adaptation and Failure Recovery by Evolutionary Computation for Multiple Heterogeneous Robots

Dr Myra Wilson, Dr Joanne Walker

EPSRC Grant (ref EP/D054419/01) EPSRC Grant EP/D054419/01

The above grant is a 3 year EPSRC funded project for one Research Associate, and a Ph.D. studentship. The project runs from Oct 2006 till September 2009. More details of the project are available on the above link.

Biologically Inspired Robotics Network

Dr Myra Wilson

EPSRC Grant (ref GR/S25340/01) http://biro-net.aber.ac.uk/

Dr Myra Wilson is the Principal Investigator for an EPSRC funded network for biologically inspired robots (biro-net). Funding runs from 01/05/03 till 31/10/06 (6 month extension granted). Biro-net aims to establish a network for advancing interdisciplinary research in the UK which will further our understanding of the underlying mechanisms that allow natural and robotic agents to adapt and survive in uncertain and dynamic environments. See the above web pages for network activities.

Artificial Immume Systems

Dr Mark Neal

Dr. Mark Neal has an interest in biologically inspired computing in general and is particularly interested in artificial immune systems. A number of pieces of work addressing a number of aspects of the immune system and potentially useful algorithms are ongoing. Drawing on immune analogies we wish to develop a multilayered system of immunised defense: (1) Innate Immune Layer: This consists of elements such as antigen presenting cells (APC) and is part of the natural immune system with which the entity is endowed at birth; (2) Adaptive Immune System: This aspect of the immune system is acquired over the lifetime of the entity and is in a continual state of change, in accordance with the environment.

Long-term Autonomy

Dr Mark Neal

Dr. Mark Neal has an on-going interest in long-term autonomy as displayed by robotic systems. The use of analogues of biological systems is the basis for his thinking and research work. In order to test the capabilities of long-term autonomous control systems he has constructed an autonomous sailing robot capable of sailing in relatively sheltered waters. The provision of locomotion through windpower means that what is typically the limiting factor for autonomous robots (energy for locomotion) is side-stepped. The robot is capable of sailing itself in a wide variety of wind conditions and currently has sufficient battery capacity to continue operation for between 7 and 10 days. Preliminary experiment's have shown promise for successful operation on this timescale, but more complete experimentation is required to test this. A number of grant proposals are either in review or preparation to construct a more robust and better instrumented version of this robot. The construction of such a robot from scratch will allow the examination of the biologically inspired systems from an unbiased starting point and provide a genuinely harsh environment (the open ocean) in which to test it.

Biologically Inspired Algorithms for Control

Dr Mark Neal

Dr. Mark Neal and a number of collaborators both within the department and at other institutions are currently developing a new more completely integrated range of biologically inspired algorithms for control. These are particularly focussed on mimicing the capabilities of biological organisms with respect to maintaining long-term autonomy: homeostasis. The three main control systems in the "higher" biological organisms are the immune, neural and endocrine systems. Biological organisms show amazing computational capabilities at many levels from individual molecules to complete organisms and populations. It is evident that these properties have been dramatically augmented during evolution, and as such they are likely to be approaching optimality in some sense. The immune, neural and endocrine systems are the three primary actors that constitute the complex cognitive system, which contributes dramatically to the ability of the higher organisms to survive in rapidly varying and hostile environments. These three systems display several major characteristics required for autonomy and cognition: self-awareness, the ability to learn, the ability to reason and the ability to deal with threats. We propose a mapping between the biological immuno-neuro-endocrine system and artificial systems that will help to solve the difficult problems involved in the construction of engineered artefacts, which are required to exhibit prolonged autonomy under analogous conditions. Such behaviour has proved unobtainable with other methods. We believe that in order to solve complex problems such as the generation of autonomous behaviour, emulation of living systems will be at least very useful and probably inevitable. The aim is to: (1) Reinterpret biological knowledge to generate appropriate computational and mathematical models of the interactions between, and key properties of the neural, endocrine and immune systems; (2) Identify and develop a framework for immuno-neuro-endocrine computation; (3) Develop and embody immuno-neuro-endocrine controllers in exemplar robotic systems; (4) Demonstrate long-term autonomy by employing these mechanisms in a hostile environment. The ultimate aim is be able to produce robot controllers that are capable of maintaining themselves in a homeostatic manner similar to that achieved by biological organisms.

Developmental Learning Algorithms for Embedded Agents

Dr Qinggang Meng, Professor Mark Lee

EPSRC grant

Currently working on an EPSRC project: "Developmental Learning Algorithms for Embedded Agents,". We are investigating a set of learning algorithms for lifelong, incremental development in embedded sensory-motor applications. The methods are inspired by developmental psychology, especially Piaget's infant cognition developmental theory with emphasis on the importance of sensory-motor interaction and staged competence learning. Infant cognition development is characterised by progression through distinct stages of competence, each stage building on accumulated experience from the level before. This can be achieved by lifting constraints when high competence at a level has been reached. Any constraint on sensing or action efficiently reduces the complexity of sensory inputs and/or actuation and provides a "scaffold" which shapes learning. During the learning process, we view errors as a mismatch in expectation, thus errors are not to be seen as exceptions but part of normal growth and experience. In our research, a hierarchical mapping mechanism is used to study the effects of different size of sensory fields and their overlapping, and investigate the staged learning from coarse to fine levels. Novelty and habituation mechanism are utilised to drive the early stage developmental learning process with some reflexes provided at the very beginning such as moving hand to mouth, and moving hand to body side rest area. Novelty attracts the system's attention and lets the system explore most interesting areas first, while habituation can cause decrease in the strength of a behavioural response to repeated stimulations, and therefore helps the system explore other areas. An experimental system has been set up to test the algorithms, it consists of two arms and one camera mounted on a pan/tilt head.

From Passive Observation to Background Awareness

Zhang, Xing (Frank)

This project introduces and concludes an exploration in developing cognitive concepts in a robot through developmental learning The main inspiration is originated from the concept of object permanence in infant development. Psychologists found that there is a relatively fixed sequence of stages on which infants develop gradually, stage by stage, to realise the concept of object permanence. In the project, a robot will develop its cognition abilities through observations of events happening in its environment. The robot will gradually develop its knowledge so as to find out the position of hidden objects. From the learning of spotting hidden objects, the robot would start to cognise the features of some special part of the environment and finally the robot would be able to know more static features of the environment and even start to cognise static objects.