R-CoRe: A Rule-based Contextual Reasoning Platform for AmI

Assaad Moawad, Antonis Bikakis, Patrice Caire, Grégory Nain, Yves Le TraonRuleML 2013 · OpenReview

Introduction

Ambient Intelligence (AmI) envisions environments—such as smart homes, hospitals, and offices—that seamlessly adapt to human needs. However, the realization of this vision is often hindered by the "imperfect" nature of context data. Sensors provide erroneous readings, batteries fail, and wireless connections drop, leading to contradictory information. Traditional logic engines are often too rigid to handle these contradictions, potentially crashing or failing to reach a conclusion when faced with conflicting data (e.g., a device reporting a user is both walking and lying down).

R-CoRe (Rule-based Contextual Reasoning) is a platform designed to address these challenges by enabling distributed reasoning in smart environments. It provides a framework for systems to make intelligent decisions even when sensors are unreliable or providing conflicting evidence.

System Architecture

The R-CoRe platform is built upon the integration of high-level logic and dynamic software management, bridging the gap between theoretical reasoning and practical distributed systems.

The R-CoRe conceptual framework integrating AAL applications, Kevoree middleware, and Contextual Defeasible Logic (CDL).

As illustrated in the architecture, the system relies on two primary pillars:

  1. Contextual Defeasible Logic (CDL): Serving as the "brain," CDL is a non-monotonic logic framework. Unlike standard logic, it can retract previous conclusions in light of new evidence. It uses mapping rules to share knowledge between different contexts and preference orderings to resolve conflicts based on trust levels.
  2. Kevoree Middleware: Serving as the "body," Kevoree is a component-based platform that supports Models@Runtime. It allows software modules to be treated as dynamic nodes that can be deployed or reconfigured on the fly, handling the "churn" of devices entering and leaving the network.

Technical Implementation

R-CoRe implements several specialized mechanisms to facilitate distributed reasoning:

  • Query Component: A custom Kevoree component representing a specific "context" (e.g., a device or a user profile) with its own local knowledge base.
  • Query Servant: A multi-threaded mechanism that attempts to solve queries locally first. If local data is insufficient, it reaches out to other nodes to gather evidence.
  • Loop Detection: A distributed tracking mechanism that prevents infinite query cycles (e.g., Node A asking Node B, which then asks Node A again).
  • Query Interceptor: A graphical tool for developers to monitor and debug the reasoning process in real-time.

Case Study: Ambient Assisted Living

To evaluate the platform, the authors simulated an Ambient Assisted Living (AAL) scenario based on a real-world network in Luxembourg. The scenario involves an 85-year-old patient, Annette, who suffers a fall.

A simulated emergency scenario where conflicting sensor data (normal pulse vs. lying on floor) must be resolved.

In this simulation, Annette's health bracelet is damaged during the fall and erroneously reports a "normal pulse." Simultaneously, an Activity Recognition Module (ARM) detects her "lying on the floor." R-CoRe resolves this conflict using a hierarchy of trust: the system is configured to prioritize the ARM and medical profile over the heart rate bracelet. Consequently, the Home Care System (HCS) correctly infers an emergency and triggers an SMS module to contact neighbors.

Insights and Results

The evaluation demonstrated that R-CoRe successfully manages decentralized reasoning without a central server. Key insights include:

  • Conflict Resolution via Preferences: Perfect data is not required if the system has a clear hierarchy of trust to resolve contradictions.
  • Dynamic Adaptability: By leveraging Kevoree, the system handles device mobility and network changes without requiring manual reconfiguration or system reboots.
  • Qualitative Success: The reasoning engine successfully translated logical conclusions into real-world actions, such as triggering emergency alerts.

Limitations and Future Work

While successful, the current implementation has limitations regarding scalability. Knowledge bases are currently capped at 500 rules, and queries have a 10-second timeout to manage limited hardware resources. Future research will focus on integrating Complex Event Processing (CEP) to better handle raw sensor streams and developing tools to allow non-experts to author logic rules using natural language.

Summary updated

BibTeX
@inproceedings{moawad2013r,
  author = {Assaad Moawad and Antonis Bikakis and Patrice Caire and Grégory Nain and Yves Le Traon},
  title = {{R-CoRe: A Rule-based Contextual Reasoning Platform for AmI}},
  booktitle = {RuleML},
  year = {2013},
  url = {https://openreview.net/forum?id=SqLEifanQlq},
}
Figure