Possible PhD Projects Associated with the Xanibot Project
[Last Updated: 8 Apr 2008]

We do not believe it is proper for the project to depend on work of PhD students, since a very important part of the experience of doing a PhD is learning how to identify a project that is worth pursuing and not too difficult. Moreover, students should have the right to change their research topics (within the broad framework of Xanibot) after learning more about the problems.

TO BE EXPANDED/CLARIFIED/MODIFIED

• Extending, or modifying some aspect of the theory/theories behind the Xanibot project

  e.g. about mechanisms, architectures, representations, requirements, nature-nurture tradeoffs.

• Modelling development of meta-configured competences

  e.g. acquiring new ways of learning based on skills previously acquired through interacting with the environment.

• Implementing aspects of the theories, using robots and simulated robots (3-D and 2-D),

• Investigation and modelling of parrots learning by play and exploration.

  Investigating the ontogenetic trajectory of physical and causal cognition in parrots,
  if possible, using parrots bred in the laboratory.

  In the longer term this might lead into a project investigating whether by providing tasks
  and problems of particular kinds, we can 'scaffold' development of unusual kinds of competences,
  compared with normally-reared birds.

• Investigation and modelling of humans learning by play and exploration.

  Task domains could be 2-D sliding puzzles, or 3-D manipulation puzzles, or open-ended 3-D manipulation 'toys',
  e.g. polyflaps.
  Effectors could be single point pushing devices, or grippers.
  Problems could include learning about what can and cannot be done by different effectors in various configurations.

• 3-D vision and understanding of affordances

  Current 3-D vision systems attempt to produce 'filled-in' surface models suitable for projecting
  views of scenes from different directions.

  What sorts of 3-D representations could be more useful for enabling a robot or animal to decide
  what actions are possible, how they might be performed, and what their consequences are likely to be?

• Acquiring and using information about proto-affordances and combinations of proto-affordances,

  Including use of epistemic affordances i.e. affordances related to information available
  as opposed to actions that are possible.

  As explained in

  http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0801
  Architectural and representational requirements for seeing processes and affordances (2008)

• Exploratory motor control in humans and robots (3-D and 2-D)

  using 2-D force controlled manipulanda to set up dynamically variable virtual visual and mechanical
  environments to test exploratory search strategies in humans.

  This may be extended to 3-D using the Phantom arm.

• Identifying scope and limits of Mosaic, H-Mosaic and alternative models.

  in collaboration between Wyatt and Miall.

• Reasoning and planning about spatial structures and processes

  recording human "point-force" manipulation of polyflaps within maze structures

• How can novel spatial interactions be predicted?

  After learning about what happens when a gear wheel is rotated on its axle,
  e.g. by pushing its teeth tangentially, how can one predict what will happen
  if two gear wheels are mounted with their teeth meshed and one is rotated?

  Compare predicting behaviours of new configurations of strings and pulleys (discussed here (1971)).

  Ontology development.

  What sorts of experiences can lead an animal or robot to realise that its current ontology is inadequate?
  How can it generate an extension to the ontology?
  How can alternative possible extensions be evaluated?

• Understanding and using causal relations of various kinds

  Kantian -- deterministic, structure-based
  Humean/Bayesian -- statistical, correlation-based
  (Discussed by Chappell and Sloman in these presentations:
  http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#wonac)

• Developing new neural models capable of supporting the above competences

  especially competences requiring compositional semantics, and recombination of separately developed modules.

• Functional brain imaging of exploratory behaviours

  using the MR compatible force controlled manipulandum to record evoked activity
  during exploration in a virtual mechanical environment, testing hypothesized self-reward for exploration

• Comparing, contrasting, and possibly integrating neural-net mechanisms
  for learning with symbolic AI mechanisms

  e.g. Case-based, or Explanation-based, learning.

• Modelling an aspect of human learning of geometry or topology by
  playing with spatial structures and noticing patterns.

  E.g. as discussed in:
  http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#math-robot

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham