System and method for an open autonomy kernel (oak)

• US 20030182354A1
• Filed: 02/14/2003
• Published: 09/25/2003
• Est. Priority Date: 06/01/2001
• Status: Active Grant

• Current Assignee: Johns Hopkins University
• Original Assignee: Johns Hopkins University
• Inventors: Watson, David p, Alger, David L, Scheidt, David H, Vick, Shon D, McCubbin, Christopher B
• Granted Patent: US 7,574,497 B2
• Time in Patent Office: Days
• US Class Current: 718/106
• CPC Class Codes: G06F 8/24 Object-oriented
• CPC Class Codes: G06F 8/313 Logic programming, e.g. PROLOG programming language
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models

Abstract

The Open Autonomy Kernel (OAK) addresses critical infrastructure requirements for next generation autonomous and semi-autonomous systems (24), including performance tracking, anomaly detection, diagnosis, fault recovery, and plant “safing”. OAK combines technologies in automated planning and scheduling, control agent-based systems (22), and model based reasoning to form a portable software architecture (26), knowledge-base, and open Application Programming Interface (API) to enable integrated auxiliary subsystem autonomy.

Claims

1. A system for controlling a connection-based system in which a plurality of independently executing control agents are distributed across an agent communication framework, each control agent being associated with a subsystem of the connection-based system wherein each control agent is responsive to other control agents such that connection-based system goals and connection-based system states can be propagated throughout the distributed network of control agents and acted upon to autonomously control the connection-based system, said system comprising:

an agent communication framework that serves as a communication network for the plurality of control agents, wherein

each control agent is comprised of;

an agent communication broker that provides a communication interface between a control agent and the agent communication framework;

a model based reasoning engine communicable with the agent communication broker that uses a first principles modeling technique to estimate subsystem states based on data gathered from subsystem hardware components and knowledge of past commands issued to said hardware components; and

a planner communicable with the agent communication broker and the model based reasoning engine that generates a plan for the implementation of a specific goal based input including the specific goal and the current subsystem state as determined by the model based reasoning engine.

2. The system of claim 1 wherein at least one control agent is communicable with an external actor that provides a system level goal to initiate connection-based system-wide control.

3. The system of claim 2 wherein a control agent is communicable with at least one hardware component within the control agent'"'"'s subsystem via a control mediator that provides an interface between the control agent and the hardware'"'"'s device controller.

4. The system of claim 3 wherein said control agent is further comprised of:

an executive manager communicable with the agent communication broker that routes incoming and outgoing messages for the control agent;

an ACL manager communicable with the executive manager for translating incoming messages into message objects, wherein message objects include a query object having data pertaining to subsystem status queries, a fact object having data pertaining to subsystem state updates, a goal object having data pertaining to the goal of a control agent, and a subscription object having data pertaining to another control agent that is interested in the status of the present control agent;

a model-based reasoning engine manager communicable with the executive manager, said model-based reasoning engine manager for storing the current state of a subsystem model; and

communicating query, fact, and goal objects between the model-based reasoning engine and executive manager;

a plan implementation manager communicable with the executive manager for implementing a plan generated by the planner;

a subscription manager communicable with the executive manager for maintaining a list of other control agents authorized to receive messages from the present control agent handling requests from other control agents and generating fact message objects pertaining to subsystem state changes to said other control agents.

5. The system of claim 4 wherein said control agent is further comprised of a reactive manager communicable with the executive manager for monitoring and comparing the current control agent'"'"'s state against a set of emergency states and a persistent goal to determine whether the current control agent'"'"'s model has entered an abnormal or hazardous state, wherein when the reactive manager determines that a control agent'"'"'s model has entered an abnormal or hazardous state, said reactive manager generates a new pre-emptive goal for the control agent to implement in order to prevent damage to the connection-based system.

6. A system for autonomously controlling a connection-based system in which a plurality of independently executing control agents are distributed across an agent communication framework, each control agent being associated with a subsystem of the connection-based system wherein each control agent is responsive to other control agents, said system comprising:

means for developing a strategy for the propagation of goals across the distributed network of control agents;

means for propagating said goals across the distributed network of control agents;

means for developing a strategy for the propagation of inferred beliefs pertaining to subsystem states across the distributed network of control agents; and

means for propagating said inferred beliefs pertaining to subsystem states across the distributed network of control agents; and

model-based reasoning means using a first principles modeling technique to perform subsystem state estimation to determine inferred beliefs pertaining to subsystem states.

7. A method of autonomously controlling a connection-based system in which a plurality of independently executing control agents are distributed across an agent communication framework, each control agent being associated with a subsystem of the connection-based system wherein each control agent is responsive to other control agents, said method comprising:

(a) assigning, via an external actor, a goal to a control agent;

(b) in the control agent assigned a goal by the external actor, developing a goal oriented plan comprised of subgoals;

(c) propagating the current subgoals to child control agents based on the subsystems within the child control agents'"'"' domains wherein the child control agents further decompose the original goal oriented plan by developing a new goal oriented plan of subgoals;

(d) repeating step (c) until the original goal oriented plan has been fully decomposed to the point where the subgoals can be translated into instructions for hardware that is within a control agent'"'"'s domain;

(e) causing a hardware component to execute an instruction set in accordance with a control agent'"'"'s plan;

(f) determining the state of a control agent'"'"'s subsystem using a model-based reasoning means; and

(g) propagating the state of a subsystem to any interested control agents that require certain subsystem state information to develop their goal oriented plan.