Model properties

[chcorbato]

From D2.2 Understanding candidate

Different types of action can be useful to increase an agent’s understanding, among them data collection, inference, model assembly, model interchange and model transformation with respect to data format, level of abstraction, and focus.

This paragraph is not fully consistent (or the consistency should be made more explicit somewhere) with the properties defined later for models. In this paragraph the model properties are:
data format, level of abstraction, and focus.

In the Model of Modelets, they are:
Modelled Concept, Locality, odelling formalism, Representation Model Class, Parent Modelets, Modelet creator, Properties(free slot to add properties)

also, later in the Requirements analysis for understanding creation, the property degree of granularity is mentioned:

An appropriate degree of granularity in the representation of Engines and Modelets, given appropriate semantic annotation, allows the Understanding Core to consider which Engines and Modelets to use in any given situation.

[chcorbato]

Regarding Modelled Concept, I think we need to make the Abox vs Tbox distinction.

For example, for a specific modelet that contains a drone trajectory, we need to represent the "referent", that is, "the current trajectory drone A is executing", or the "planned trajectory of drone B" (Abox). We also need to represent that this "referent" is of type "trajectory" (Tbox).
This is closely related to the now footnote about the Modelling formalism, but in an orthogonal modelling dimension (modelling language in the later, modelling content in the former)

I think the Parent Modellets refers to the Tbox information about a modellet. E.g. there are models that are "estimations", they can be obtained by multiple "process models", such as "averaging values for the same reference".

[chcorbato]

More about "Parent Modellets"
I think we need to model that a modelet is the output of something, that something not being an engine, but a concrete instance of exertion of an engine with a certaint set of input modellets. This can be modelled of an instance of an exertion relationship that connects inputs with engine with output(s)

[alex-quin-gabriel]

I think we need at least 3 'dimensions' to describe modelets, with potentially additional properties

• the world model: 'a shoe' - red, size 43
• the modelling model: 'a visual representation, 2d projection' - from the top
• the engineering model/formalism: 'an rgb image' - resolution, color space

this leaves out the perception circumstances

• camera-object distance, lens distortion,

what do we need, if we want the agent to also be able to understand e.g. the amazon picking challenge reflection problem?

[chcorbato]

what do we need, if we want the agent to also be able to understand e.g. the amazon picking challenge reflection problem?

We need to list cocnrete understanding cases in the Testbeds and analyze what we need to represent from there, keeping generality in the horizon (not fogeting about it).

This is the new architecture process approach now reported in D2.2. :-)

[alex-quin-gabriel]

We need to list cocnrete understanding cases in the Testbeds and analyze what we need to represent from there.

I agree. I propose we use the previously identified scenarios to do that. We selected them for this reason (to have a limited set of concrete problems to drive architecture development) but then stopped to continue working on them, before identifying concrete cases.

[alex-quin-gabriel]

to make that work though, we need detailed enough descriptions of the non-coresense parts of the respective systems...otherwise I fear the understanding problem definitions will not be detailed enough to be useful.

[chcorbato]

Agree, @rsanz @cgs-AI maybe together with the database of requirements we need an accompanying repository of scenarios where we keep the most up to date versions. Maybe a repo on github with sysmlv2 files per snecario and accompyng tabular descriptions and visualizations? (non SysML ones, just for communiaction purposes)