Simulation of Adaptive Behaviour, Spring 2005
Course developed by: John Bird, Phil Husbands, Matt Quinn, Anil Seth, Emmet Spier and Adrian Thompson

Emmet Spier. Room 5C13 x3594. emmet@cogs.sussex.ac.uk

Assessment: Students are expected to contribute fully to each seminar and to write an essay of at least 3000 words on a topic related to material covered during the term. The essay is to be handed in the first day of summer term.

Course outline [pdf]

Online Course Library collection of digitally available course readings

Assessed Essay Details and Tips


* Introductory Lecture

An introductory lecture (Adrian Thompson) [link]

Reading

Brooks, R (1991) Artificial life and Real Robots, Proceedings of ECAL91, 3-10. A short overview, available from the course library
Jakobi, N (1998) Evolutionary Robotics and the Radical Envelope of Noise Hypothesis, Adaptive Behavior, 6(2), 326-368. A detailed example, available from the course library
Webb, B (in press) Can Robots Make Good Models of Biological Behaviour? Behavioural Brain Sciences A long, less active, overview


* 'Ancient SAB' Seminar

Reading

Available from Celia McInnes in the COGS library

Grey Walter (1950) An Imitation of Life, Scientific American, 182(5).
Grey Walter (1951) A Machine That Learns, Scientific American, 185(2).
Pask, Gordon (1962) A Proposed Evolutionary Model. In `Principles of Self- Organization', Von Foerster and Zopf (Eds), 299-253, Permagon Press.

-- Last year I convinced myself that the Pask paper was a cellular automata model similar to Conway's game of life. Is this believable given his details?

Some (old) links

Grey Walter's robots rebuilt
Reference to Owen Holland paper about Grey Walter and the robot rebuild.
Gordon Pask information.
From Complexity to Perplexity on-line version of Sci. Am. critique of `The Science of Complexity' (original magazine article is longer)
Randall Beer Proponent of the dynamical systems approach. See "On the dynamics of small continuous-time recurrent neural networks" cf. Grey Walter's "two neurons"
MS Windows simulation of Walter's tortoises.
More tortoise stuff.


* Fungus Eaters' Seminar

Pfeifer (1996) Building "Fungus Eaters": Design principles of autonomous agents. SAB98 pp 3-12 (in on-line library)
Q: Can the principles be ranked in terms of their importance for building robots? What is/are the most fundamental for adaptive behaviour?

Stein and Meredith (1993) The merging of the senses. Ch.2. QU 4590ste. (copies from COGS library set)
Q: What is the major difference between the sensory systems of unicellular organisms, such as the protozoan Paramecium and vertebrates?

Rucci, Wray, Tononi, Edelman (1997) A robotic system emulating the adaptive orienting behaviour of the barn owl. Proc. 1997 IEEE Int. Conf. on Robotics and Automation, Albuquerque, New Mexico. pp 443-448. (copies from COGS library set)
Q: Assess the robot in terms of Pfeifer's design principles. Which have been implemented? What are the advantages of this particular multimodal architecture?

Agah, de Garis, Korkin, Simihga and Cho (1997) Architectural and functional specifications for a robot kitten "Robokoneko" to be controlled by a 10,000 evolved neural net module artificial brain. (also part of COGS library set) [Link] [Robokoneko Images]
Q: Assess the "world's first artificial brain" in terms of Pfeifer's design principles. Which have been implemented? How would you characterise the control architecture? Assess Korkin's (New Scientist, 9-1-99) "What is do special about this neural network is a much higher degree of biological relevance" with reference to the Stein and Meredith chapter.


* Action Selection Seminar

All of the readings selected this week show different way to build systems that have multiple goals/tasks -- this is where the problem of action selection arises. Maes wrote an influential paper (her history was in Expert systems and, like Pfeifer and Steels, the, in part, rejection of them) -- she provides some definitions to think about. Scheier and Lambrinos worked in Pfeifer's lab and produced an excellent example of robot work, how does their work reflect Maes's. Seth, from Sussex, philosophically rejects (what did you expect?) the divisons that action selection places on a system. Blumberg considers action selection using ideas from the study of animals. Humphries uses reinforcement learning, is this formalism the way to go? And, if you haven't read it, the Brooks paper give the down and dirty on the subsumption architecture -- technically not action selection but why not?
All papers available from the SAB digital library -- read them with an eye to the various stategies to sequence behaviour between mutiple tasks.

Maes, P. (1991) A bottom up mechanism for behavioural selection in an artificial creature. Proc. SAB 90, pp 238--246
Scheier, C. and Lambrinos, D. (1996) Categorization in a real-world agent using haptic exploration and active perception. Proc SAB96, pp 65-74.
Seth, A. (1998) Evolving action selection and selective attention without actions, attention or selection. Proc SAB 5, pp 139-147
Blumberg, B. (1994) Action-selection in HAmsterdam: Lessons from ethology. Proc SAB94, pp 108-117.
Humphrys, M. (1996) Action-selection using reinforcement learning. Proc SAB96
Brooks, R. (1985) A robot that walks: emergent behaviors from a carefully evolved network. MIT AI Memo 1091


* Animals as Cost Based Robots Seminar

I promise, much less to read this week.

McFarland, D (1992) Animals as Cost-based robots. International Studies in the Philosophy of Science, 6, 133-53. Reprinted as Chapter 6 in "The Philosophy of Artificial Life" M. Boden (Ed), OUP 1996. Available from Celia in COGS library.

This paper looks like it may be missing the point, even so it makes some fundamental points.

Some questions to think about:
1. Is the frame problem a problem?
2. Do all robots have niches?
3. Are McFarland's type of cost functions sensible; are they related to ones used in GAs?
4. How autonomous should a robot be?
5. When should robots use implicit, embedded or explicit control systems?
6. Why distinguish between a cost and goal function. Could goal functions be learnt?


* Evolution of Communication Seminar

See this [file].


Now is a good time to start thinking (if you haven't already) about an essay topic. We will have a seminar where you can discuss and further forumulate your ideas in a few weeks time.

* An Engineering Spin-off: 'Ants'

See this [file].


* Essay discussion

As discussed in our last seminar the next seminar will comprise of short ( < 10 minute) presentations and then discussions of a topic that you think you would like to write about for the assessed essay.

There will be no (additional) reading for this seminar.


Worthwhile links
Please send me any you find

Ezequiel's superb collection of links to on-line papers.
EASy references repository
Some of Rodney Brooks' papers
Downloadable textbook on Adaptive Control
Dario Floreano's on-line papers
Inman Harvey's page with most Sussex evolutionary robotics stuff. SAB2002conference
Julie Rutkowska: relationship between infant psychology and Alife.