Academic Course:11 Systems with Internal Models

  • 185 views
Uploaded on

By Alan Winfield

By Alan Winfield

More in: Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
185
On Slideshare
0
From Embeds
0
Number of Embeds
3

Actions

Shares
Downloads
0
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Designed by Alan Winfield Self-Awareness in Autonomic Systems Systems with Internal Models
  • 2. Designed by Alan Winfield Outline • Why are internal models important? • What is an internal model? • Examples from robotics • A visual abstraction • The major challenges of internal modes – The internal representation – The reality gap – Connecting the internal model – Making it work • Towards a generic architecture for internal models
  • 3. Designed by Alan Winfield Why do self-aware systems need internal models? • Because the self-aware system can run the internal model and therefore test what-if hypotheses* – what if I carry out action x..? – of several possible next actions xi, which should I choose? • Because an internal model (of itself) provides the self in self-aware *See Dennett’s Tower of ‘generate and test’ in Dennett, D. (1995). Darwin’s Dangerous Idea, Penguin.
  • 4. Designed by Alan Winfield What is an internal model? • It is a mechanism for representing both the system itself and its current environment – example: a robot with a simulation of itself and its currently perceived environment, inside itself • The mechanism might be centralized (as in the example above), distributed, or emergent
  • 5. Designed by Alan Winfield Examples • Examples of conventional internal models, i.e. – Analytical or computational models of plant in classical control systems – Adaptive connectionist models such as online learning Artificial Neural Networks (ANNs) within control systems – GOFAI symbolic representation systems • Note that internal models are not a new idea
  • 6. Designed by Alan Winfield Examples 1 • A robot using self- simulation to plan a safe route with incomplete knowledge Vaughan, R. T. and Zuluaga, M. (2006). Use your illusion: Sensorimotor self- simulation allows complex agents to plan with incomplete self-knowledge, in Proceedings of the International Conference on Simulation of Adaptive Behaviour (SAB), pp. 298–309.
  • 7. Designed by Alan Winfield Examples 2 • A robot with an internal model that can learn how to control itself Bongard, J., Zykov, V., Lipson, H. (2006) Resilient machines through continuous self-modeling. Science, 314: 1118-1121.
  • 8. Designed by Alan Winfield Examples 3 • ECCE-Robot – A robot with a complex body uses an internal model as a ‘functional imagination’ Marques, H. and Holland, O. (2009). Architectures for functional imagination, Neurocomputing 72, 4-6, pp. 743–759. Diamond, A., Knight, R., Devereux, D. and Holland, O. (2012). Anthropomimetic robots: Concept, construction and modelling, International Journal of Advanced Robotic Systems 9, pp. 1–14.
  • 9. Designed by Alan Winfield Examples 4 • A distributed system in which each robot has an internal model of itself and the whole system – Robot controllers and the internal simulator are co- evolved O’Dowd P, Winfield A and Studley M (2011), The Distributed Co-Evolution of an Embodied Simulator and Controller for Swarm Robot Behaviours, in Proc IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2011), San Francisco, September 2011.
  • 10. Designed by Alan Winfield A visual abstraction Maturana, H. R., & Varela, F. J. (1987). The Tree of Knowledge: The Biological Roots of Human Understanding. Boston, MA: New Science Library/Shambhala Publications.
  • 11. Designed by Alan Winfield …for a self-aware artificial system A self-aware machine From lecture by Prof Roger Moore: Extending Maturana& Varela’s symbols, FECS, Feb. 2012.
  • 12. Designed by Alan Winfield Major challenges 1 • To model both the system and its environment with sufficient fidelity, including: – The system and its behaviours – The world and its physics – The system’s sensorium – The effect of interactions between the modelled system and modelled world, on its modelled sensors • But what is sufficient fidelity?
  • 13. Designed by Alan Winfield Major challenges 2 • Example – imagine placing this Webots simulation inside each NAO robot: Note the simulated robot’s eye view of it’s world
  • 14. Designed by Alan Winfield Major challenges 3 • The Reality Gap – No model can perfectly represent both a system and its environment. Errors in representation are referred to as the reality gap. • The effect of the reality gap will likely be to reduce the efficacy of the system’s self-awareness and therefore its ability to accurately model unexpected events or possible actions – But this is speculation – it’s an open research question
  • 15. Designed by Alan Winfield Major challenges 4 • To connect the internal model with the system’s real sensors and actuators (or equivalent) – i.e. so that events in the real world, as sensed by the system, are represented in the model • To synchronize updating the internal model from both changing perceptual data, and efferent actuator data – i.e. so that the internal model in in-step with the real system and its environment • Except when the model is being used to test what-if hypotheses
  • 16. Designed by Alan Winfield Major challenges 5 • Making it all work – Building internal models that • represent a system and its environment, • hooking the model up to the system’s perception and actuation system, • making use of the model to moderate behaviour (i.e. for safety), and • smoothly integrating all data flows to make it all work – is immensely challenging and remains a difficult research problem
  • 17. Designed by Alan Winfield A Generic Architecture • The major building blocks and their connections: Control System Internal Model Sense data Actuator demands The loop of generate and test The IM moderates action- selection in the controller evaluates the consequences of each possible next action The IM is initialized to match the current real situation
  • 18. Designed by Alan Winfield Conclusions • Would such a system be self-aware? – Yes, but only in a minimal way. It might provide sufficient self-awareness for, i.e. safety in unknown or unpredictable environments • But this would have to be demonstrated by the robot behaving in interesting ways, that were not pre- programmed, in response to novel situations • Validating any claims to self-awareness would be very challenging http://alanwinfield.blogspot.com/