Continuous Case-Based Reasoning

1
Continuous Case-Based Reasoning

• Ashwin Ram and Juan Carlos Santamaría
• (1993)

2
What is it about?

• Investigates the use of CBR in continuous,
  real-time problem domains

3
Some characteristics of such domains

• Require continuous representations
• Require continuous performance
• Require continuous adaption and learning

4
Application domain

• Robot navigation (selecting robot behaviours)
• Real-time
• The domain is continous (constant flow of data
  from the robots sensors)

5
Schema-based reactive control

• Schema-based navigation is a form of reactive
  control that decomposes actions into a set of
  multiple concurrent processes called motor
  schemas. Each of these active behaviors generates
  a velocity vector based upon immediate sensory
  data which is combined with the outputs of the
  other behaviors. The result is transmitted to the
  robot for navigation.

6
(No Transcript)
7
(No Transcript)
8
ACBARR

• A Case-BAsed Reactive Robotic (system)
• A case contains
• Assemblages of behaviors
• Behavior adaptions

9
(No Transcript)
10
(No Transcript)
11
Moving on...

• ACBARR relied on a fixed library of handcoded
  into the system
• SINS (Self-Improving Navigation System) can learn
  new cases through experience

12
SINS
A case in SINS consists of associations between
sensory inputs and schema parameters
13
(No Transcript)
14
(No Transcript)
15
Assumptions

• Environment is
• Causal same actions executed under the same
  environmental would result in same/similar
  outcomes
• Consistent small changes in executed actions -gt
  small changes in actions

16
Assumptions

• Experiences are likely to be typical of future
  experiences
• Problem domain can be represented quantitatively

17
Innovations

• Continuous cases
• Abstract cases
• Virtual cases
• Two types of behavior modification
• Two types of adaption
• On-line real-time response
• Adaptive reactive control

18
Conclusions

• Open issues
• Size and extent of cases
• Cases are not easy interpret