Key focus of conference was on grounded artificial life research. Main contributions in the
embodiment of a-life in both real robots and simulated robotic agents.

Day1
Monday 13th September
Tutorials
 
 Six Tutorials were run in two parallel sessions

    Session 1
(T1)Artificial Chemistries     P Ditttrich
(T2)From a four character code to a living organism: the challenge of genomics.    C.V. Jongeneel
(T3)Cellular Automata and self-replication     M. Slipper

    Session 2
(T4)Evolutionary Robotics         Stefan Nolfi.
Stefan Nolfi is working in the area of  developing autonomous robots through a self-organising process based on artificial evolution. Dario Floreano and Stefan Nolfi have showed how Co-evolution (Evolving multiple populations in concurrent generations) can produce incremental evolution without requiring additional supervision. Successive generations of both Prey and Predators had increased complexity compared to evolving single populations against "Best Individuals". Complications arise for co-evolution stratagies. (Prey changes strategy => Old predator strategy ineffective). Individuals in the population are seen to be plastic and the co-evolution of best prey/predator can evolve best prey AND best predator.
 
(T5)An introduction to Ant Algorithms     Marco Dorigo
Marco Dorigo gave an interesting and stimulating presentation on how indirect social communication (Stigmergy) can be employed artificially in Ant Algorithms to provide solutions to classical computational problems. e.g. Shortest Path, Clustering, Maximum Demand.  One Ant Algorithm employs biologically inspired pheromone deposition to Positively Re-inforce the shortest path route. i.e. The more Ants that have taken a particular route, the higher the probability that subsequent Ants will take that particular route. Applications of ant algorithms are, for instance, internet routing, task allocation, and sorting. Ant Algorithms have been tested against Travelling Salesman Problem, Quadratic Assignment Problem, Sequential Ordering Problem etc. with as-good-as or better-than status against GA's and other Algorithmic methodologies.

(T6)Autonomous Virtual Humans        Daniel Thalmann
Daniel Thalmann captivated those present with some simulations of human populations which had both collective and individual characteristics. The virtual simulations of "crowds" in motion gave a visual insight into the problems of flocking, convergence and dispersal when used to simulate real motion in population of individuals. Differences between the characteristics of "political demonstration" type crowds and "soccer supporter" type crowds may seem easy to visually differentiate but algorithmically pose quite a challenge.

Day 2
Tuesday 14th September
Epistemology - Evolutionary Dynamics - Evolutionary Cybernetics I

Richard Watson gave a talk on how symbiosis can guide evolution where individuals in a population which forge a symbiotic relationship to find a solution. This non-genetic mechanism is used to guide the genetic make up of organisms by shaping the evolutionary landscape. he demonstrates how symbiotic scaffolding (inter generation alliances) can guide the genetic make up of organisms and evolve organisms that would not otherwise have evolved.

    Sevan Ficici gave demonstrations and a talk on Pursuit/Evasion strategies. By using statistical tools to analyse evasion behaviour, it was possible to create a hand built optimised pursuer (to react to complex evasion behaviour). Evolution against this pursuit strategy resulted in complex evasion behaviour. Further work will involve the use of these statistical tools to track coevolutionary progress.

    On Self-Reproduction and Evolvability, Tim Taylor describes Von-Neumann's self-reproducing architecture and argues that a-life platforms such as Tierra reproduce genetically rather than by self-inspection. He states that self-reproduction does not imply high evolvability. Implicit V's Explicit encoding of automata are discussed and questions arise as to what is part of the architecture of the machine and what is part of the physical environment (Operating System).

Day 3
Wednesday 15th September 
Evolutionary Cybernetics II - Bio-Inspired robotics and autonomous agents - Self-replication, self-maintenance, and gene expression

    Ranit Aharanov-Barki demonstrate the use of structure called a "command neuron" in an autonomous agent controlled by an evolved Artificial Neural Network (ANN). The agents exhibit two distinct behaviours, exploration and foraging, and the "command neuron" translates the sensory input into a binary command which modulates the dynamics of the ANN and thus switches between the two behaviours. This gives an overall effect of dynamic policy management in the autonomous agent.

    On the Dynamics of Robot Exploration Learning was the title of the talk by Jun Tani. Confusion (internal modeling) in Real Robots when in exploration mode is shown to be a good thing. It develops diverse exploratory behaviour in the robot which results in a more rational model of the environment in the end.

    A talk on Brachiation Robot Controller by Yasuhisa Hasegawa was animated in video to show the ape-like robot attempting to swing from one branch to another. Using behaviour controllers on the actuators of the locomotion robot the overall behaviour of robot is changed when the parameters are modified (e.g. branch distance). The video of the swinging robot (on success) brought a load applause from the audience, a true reflection on the complexities of such a control/ prediction task.

Day 4
Thursday 16th September
Societies and collective behaviour I & POSTERS

    The main focus of the day for me was on Luc Steels  talk on Cognitive Teleportation and Situated Embodiment. Luc, who is working at the SONY Computer Science Lab in Paris, said that the motivation for his work was to understand the origins of intelligence using biological principles. Intelligence was said to be not just sensory-motor intelligence but also communication. This requires categorisation of reality, complex processing and enables cultural storage of knowledge. It is important therefore that we do not lose sight of embodiment. There is a basic need to understand the origins of complexity and evolution in living systems, Luc states that language can be seen as a living system. A historical view of the Stages of Human Language was given from material to grammar and a list of contributions from MacLennon, Werner and Dyer, Hurford, Oliphant, Billard, Cangelosi, Batali, Kirby and Steels were cited before the Talking Heads experiment was introduced. This experiment in conjunction with Frederic Kaplan involves Two Robot Agents( from a population of hundreds ) looking at the same image through two steerable cameras. The image is segmented and a random selected portion of the image is identified by the First Agent. The First Agent then conveys this portion (Speaks) to the Second Agent. The Second Agent guesses which portion of the image ( which it also has segmented) is being alluded to and if correct it expands its vocabulary to include that utterance. This experiment in the origins of lexicon (meaning) shows that a lexicon/ontology does not need to be programmed but can evolve, it also shows that this lexicon can be self-regulating and handle multiple references to similar/same objects but also provides damping of synonymy (multiple words for same object) and resolves polysemy (ambiguity). It is possible to track semiotic dynamics through the lexicon as "words" go from vague to precise meanings. New agents join the game all the time which keeps the system more regular, these inexperienced agents have to invent words when they have no lexicon of their own and talk to more experienced agents which may have a comprehensive lexicon. The best thing about this experiment is that everyone can participate, so why not join Talking Heads and Launch your own Agent ....

 Jason Noble was talking about Rats.... to be specific An evolutionary Model of Social Learning about food by Norway Rats. Jason talked about two ways that this species learn from one another. Firstly, imitation, they copy the feeding preferences of their conspecifics when they smell a novel food odour on the conspecifics breath even above it's normal diet. Secondly, they will spontaneously follow conspecifics out of the nest on foraging trips with a view to finding a new food source. Despite their social skills they don't seem to be able to learn aversions after seeing a sick conspecific. A model is created for this behavior in environments of high and low food toxicity and the outcomes discussed.
 

Day 5
Friday 17th September
Societies and collective behaviour II
 
Collective Learning and Semiotic Dynamics was presented by Frederic Kaplan where more specific detail on the Talking Heads project was introduced. Frederic conveys the hypothesis that language is indeed an evolving complex dynamic system which self-organises and gets transmitted in a cultural process. The analysis of semiotic dynamics in the lexicon of the agents shows the through adaptive nature of the language as a system.

    Simon Kirby gave a memorable presentation on Syntax out of Learning. The paper examined social/cultural transmission of language implications to the origins of syntactic structure. Induction algorithms use token/meaning pairs as input to produce grammars. New categories with their own rules are invented and added to the grammar by techniques within the induction algorithm such as "chunking". The paper discusses other influences that this approach has including Emergence of Vocabulary and Emergence of Recursion. Simon shows how linguistic evolution can occur without biological evolution in an unseeded environment.
 
A number of biologists gave keynote lectures : Meinhardt, Hamilton, and Lenton
APOLOGIES - To presenters that I did not see and have not commented on.
                         There was a heavy schedule with a lot of presentations, posters and demonstrations,
                            (SO MUCH TO SEE so little time)

 
Day 6
Saturday 18th September EWLR-8
Eighth European Workshop on Learning Robots

Excellent fluid workshop on Evolutionary Robotics, Embodied Reinforment Learning and Control Architectures.
This workshop was chaired by Jeremy Wyatt and the guest speaker Henrik Lund, Aarhus University, Denmark
who is head of the Lego Lab gave an insight into the flexibility and evolvability of  Mindstorm . Currently, the LEGO Lab has approximately 10 different projects running, which all concern evolutionary robotics and adaptive robotics. Development work where children will decide the learning methodology (graphically) of their robots sounds both inspired and exciting.

Akio Ishiguro gave a talk on Dynamically-Rearranging Neural Networks which employed bio-inspired neuromodulator technology to facilitate switching of learning types. This is similar in concept to a hierarchical reinforcement learning net (e.g. Hierarchical Q-l or W-l nets) which performs action selection (behavior/policy wise) as in ECAL99 participant Ranit Aharanov-Barki's talk on "command neurons".

(for more info see Henrik Lund report)