Skip to main content

A RDFlib-backed minimalistic knowledge based for robotic application

Project description

KnowledgeCore is a RDFlib-backed minimalistic knowledge base, initially designed for robots (in particular human-robot interaction or multi-robot interaction).

It stores triples (like RDF/OWL triples), and provides an API accessible via a simple socket protocol.

pykb provides an idiomatic Python binding, making easy to integrate the knowledge base in your applications.

It has almost no features, except it is fast and simple. Basic RDFS reasoning is provided.

Written in Python. The only required dependency is sqlite3. If rdflib is also available, you can easily import existing ontologies in RDF/OWL/n3/Turtle formats in the knowledge base.

Documentation

You can use KnowledgeCore either as a server, accessible from multiple applications (clients), or in embedded mode (which does not require to start a server process, but is limited to one single client). Note that the embedded mode is only available for Python applications.

In both case, and if your application is written in Python, it is highly recommended to use pykb to interact the knowledge base.

Server mode

To start the knowledge base as a server, simply type:

$ knowledge_core

(run knowledge_core --help for available options)

Then:

import kb

with kb.KB() as kb:
    #...

See usage examples on the pykb page, or in the KnowledgeCore unit-tests.

Embedded mode

No need to start KnowledgeCore. Simply use the following code to start using the knowledge base in your code:

import kb

with kb.KB(embedded=True) as kb:
    #...

Interacting with KnowledgeCore from other languages

  • from C++: check liboro

  • from any other language: the communication with the server relies on a simply socket-based text protocol. Feel free to get in touch if you need help to add support for your favourite language!

Features

Server-Client or embedded

KnowledgeCore can be run as a stand-alone (socket) server, or directly embedded in Python applications.

Multi-models

KnowledgeCore is intended for dynamic environments, with possibly several contexts/agents requiring separate knowledge models.

New models can be created at any time and each operation (like knowledge addition/retractation/query) can operate on a specific subset of models.

Each models are also independently classified by the reasoner.

Event system

KnowledgeCore provides a mechanism to subscribe to some conditions (like: an instance of a given type is added to the knowledge base, some statement becomes true, etc.) and get notified back.

Reasoning

KnowledgeCore provides RDFS/OWL reasoning capabilities via the reasonable reasoner.

See reasonable README for the exact level of support of the different OWL2 RL rules.

Transient knowledge

KnowledgeCore allows to attach ‘lifespans’ to statements: after a given duration, they are automatically collected.

[this functionality is currently disabled. Please open an issue of you need it urgently]

Ontology walking

KnowledgeCore exposes several methods to explore the different ontological models of the knowledge base. It is compatible with the visualization tool oro-view.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

KnowledgeCore-2.6.0.tar.gz (29.7 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page