Общие сведения
Архитектуру HOMER [VB90] разработали для автономных подводных аппаратов в 1990 г. Стивен Вер (Steven A. Vere) в Лаборатории реактивного движения НАСА, (который также разработал планировщик DEVISER - первый планировщик для НАСА на базе ИИ), и Тимоти Бикмор (Timothy Bickmore), который сейчас продолжает работу над когнитивными архитектурами в Северо-Восточном Университете (Northeastern University), Бостон, Массачусетс, США.
Bickmore's Personal Homepage
Now Professor Bickmore is interested in the development and study of relational agents, computer agents designed to build and maintain long-term, social-emotional relationships with people. In order to use the same myriad cues that people use when relating to each other, he builds agents that are capable of emulating face-to-face interaction with people, including the use of hand gestures, facial expressions, and body posture, in addition to speech.

Lab: Relational Agents Group | Current Research Projects
The Relational Agents Group is a research laboratory in the College of Computer and Information Science at Northeastern University
Relational Agents are computer agents designed to form long-term, social-emotional relationships with their users. We are investigating the use of these agents in task domains in which human-agent relationships actually improve task outcomes, such as in coaching, counseling, psychotherapy and healthcare.

---

Homer
Description
The Basic Agent is an attempt to create a conscious, mind-like AI artifact embedded in time and functioning in a simulated dynamic environment.

The agent integrates temporal planning, temporal reasoning, reactive replanning, action execution, limited natural language understanding and generation, symbolic perception, episodic memory and reflection, and some general knowledge associated with the vocabulary items.
A state-change style semantics for action verbs, based on a set of primitive relations, allows convenient integration of the natural language and planning components.

Homer [VB90] is an integrated AI artifact which is embedded in a simulated, single-agent, object-cluttered marine environment called the Seaworld.
Homer integrates a number of deliberative capabilities including limited natural language generation and recognition (using an 800-word English vocabulary and medium coverage grammar), temporal planning and reasoning, acting on and perceiving its environment, as well as the ability to reflect upon its own experiences.

Homer’s operations are centred around a temporal planner which it uses to synthesise plans in response to a human user’s natural language goal commands.
Goal commands typically include time constraints on their achievement and preservation. The planner then imposes goal protection conditions which are constantly monitored in order to detect plan violations, both at plan time or later during execution. Associated with the planner is a set of declarative activity models in precondition/postcondition format, describing all of the actions, inferences, and events that the agent knows about.

Homer employs one of these actions — the “go” action — to plan collision-free trajectories to desired locations. Besides being able to form and retain compound future plans, Homer’s planner is also capable of limited replanning in order to accommodate additional goals that are given to it.

Homer sets out with limited knowledge of its world, eventually gaining new information about the different objects in its world either by perceiving them or by being told about them directly by the user. Homer time-stamps and records all events which have transpired in its life by placing these in its episodic memory. Homer’s events include its perceptions, all of the goal commands and questions it has previously received from the user, as well as any actions that it has taken. With the use of its temporal planner and various reflective processes for monitoring and processing its personal memory, Homer is able to make inferences and provide answers to a range of questions about its past experiences, present activities and perceptions, as well as its future intentions.

Homer's Architecture was not designed for general intelligence.
Its underlying philosophy was the synthesis of several currently distinct areas of AI to form one complete system.
Vere and Bickmore believe that the research in the fields of planning, learning, natural language understanding, robotic navigation, etc. had progressed sufficiently to allow the creating of a useful architecture and agent.

HOMER is modeled after an unmanned submarine. It exists in a simulated two-dimensional sea world. The world also has temporal extent. The agent's actions and external events in the world are distinguished by their coordinates in a plane and the time at which they take place.
A graphical representation of this world is displayed on a terminal to fascilitate interaction with the user.
HOMER's environment contains a variety of objects one might expect to find in an aquatic setting; fish, birds, boats, piers, buoys, islands, and other agents.

The architectural elements of the system and their interaction are described below:

Architecture:
The architecture on which Homer exists is a modular architecture.
It consists of a memory, a planner, a natural language processor, some monitor processes, and a plan executor.
This modular design is due to their ambition to integrate several research into several AI components into a single working agent. In the future, they plan in include an inductive learning module.
There is no true central control module, although one might argue that the planner fulfills most of this function.

The memory can be divided into two sections, general knowledge, including world knowledge and lexical knowledge, and episodic knowledge. The planner is interesting is that it is a temporal planner. This planner as well as its associated reasoner are derived from DEVISER V planner.
The natural language processor accepts input from a human (via keyboard) parses it, and outputs results to a screen using a sentence generator. Lexical knowledge is contained in the general memory.
The plan executor works closely with the monitor process to attempt successful completion of the plan.
The monitor, or reflexive, processes provide feedback to the plan executor in case replanning is necessary.

Agent Properties
One of the most interesting properties of this architecture is its temporal extent. This is due to its episodic memory and its temporal planner. The knowledge base in Homer is global. Any of the modules can access it, although only the input modules, such as the text interpreter and world sensors, and the reflective processes can modify it.
Knowledge is represented in a frame-like manner.
Knowledge consistency is handled by simply stating that the most recently learned fact is the true one.