Strategy Generation with Planning

Researcher: Thomas Heider

Runtime: 4/2000 - 4/2003

The Smart Appliance Lab operates in the context of users who have intentions. The system should automatically realise the users intention on base of the existing components. The user should perceive the complete set of (networked) devices as a single system of interoperating components that helps him in reaching his specific goals. In order to support this kind of interaction, we need a system that is able to reason about the effects a device operation has with respect to the environment state as perceived by the user. Therefore each component has to describe its functions ability and its effects on the environment by using a suitable definition language. Based on such a formal functionality definition, planning algorithms are able to create strategies to fulfill the users goal by using the available device infrastructure, on base of the operators offered by these infrastructures.

As example, consider the situation outlined in the left Figure, where a user would like to increase the brightness of his TV set. Assuming the TV is already set to maximum brightness, the sensible reaction of the ensemble would be the one given at right: reduce ambient light. In order for an ad hoc ensemble to arrive at this conclusion, TV set, lamp, and shutter must provide a description of their capabilities, similar to the one given below. (For sake of brevity, this capability definition has been very much simplified.)

The Lamp’s impact on the Environmentstate:

Action: dim-down(?x)

Precond: luminosity(?x) = high

Effect: luminosity(?x) = low

 

 

The Shutter’s impact on the Environmentstate:

 

Action: closeShutter(?x)

Precond: open(?x)

Effect: ¬open(?x) ∧ luminosity(?x) = low

 

 

The TV’s dependance of the Environmentstate:

 

Axiom: ambBrightness-low

Context: ∀?x ∈ dom luminosity : luminosity(?x) = low

Implies: ambientBrightness = low

 

Axiom: increaseTVBrightness(?x)

Context: brightness(?x) = max ∧ ambientBrightness = low

Implies: brighter(?x)

 

Then, based on a specific situation given by

Inits: (brightness(TV) = max ∧ luminosity(Lamp)= high ∧ dimmable(Lamp) ∧ open(Shutter))

 

a suitable plan for the Goal:

brighter(TV)

 

could the be computed as Plan:

 

[dim-down(Lamp), closeShutter(Shutter)].

Key publications

Thomas Heider and Thomas Kirste. Multimodal appliance cooperation based on explicit goals: concepts & potentials. In sOc-EUSAI '05: Proceedings of the 2005 joint conference on Smart objects and ambient intelligence, pages 271-276, New York, NY, USA, 2005. ACM Press.

Thomas Heider. Goal-oriented assistance for extended multimedia systems and dynamic technical infrastructures. In 7th IASTED International Conference on Internet and Multimedia Systems and Applications 2003, Honolulu, USA, August 2003.