Our Technology

Summary: PhysarAI is different. It actively pursues goals. It doesn’t just perform tasks or mimic patterns from large data sets. It understands what needs to be done and coordinates networks of people and devices. We call this approach advanced ambient assistance. It means your environment actively works to support your goals. It’s a potential game-changer in many industries.

​

Overview

​

PhysarAI is an adaptation of Goal-Oriented Action Planning (GOAP) - a technique traditionally used to animate non-player characters in video games. It's a smart AI system designed to make decisions based on specific goals within set environments. The main focus of PhysarAI is on managing goals, abilities, and planning and executing actions based on what's currently happening in its representation of the world.

One of the standout features of PhysarAI is its planning method, which works backward from the goal to the current state of the world. This is especially useful in places where the goals and variables are clear and well-defined. On the other hand, Large-Language Models (LLMs) tend to make predictions based on data they receive, which is flexible, but can be less reliable when you need specific outcomes in controlled settings.

PhysarAI is particularly skilled at managing and prioritizing goals. It aligns the system’s abilities with the desired results. While LLMs are very flexible, they usually need extra programming or external systems to help them focus on and manage specific goals.

Design

PhysarAI interacts with a model of the "world", which represents the current state of the environment as a list of name-value pair tuples. Each part of this world has specific characteristics and values. The AI sets goals as desired states in this world, choosing which goals are most important based on how urgent they are and continuously checks if these goals have been met.

The system’s abilities are the actions it can perform. Actions are represented as object-oriented implementations of a standard interface. Actions conforming to the interface must be fully implemented and registered with the system to be used in planning. Each action is registered as an ability that has conditions and effects that dictate when it can be used and how it changes the world.

Using its backward planning method, PhysarAI creates a sequence of actions to achieve a chosen goal. This planning is based on the current world state and the AI’s abilities. If a good plan is made, PhysarAI carries out the actions one by one, checking each action’s effect to see if it moves closer to the goal.

Execution

In a typical cycle, PhysarAI starts by picking a goal that hasn’t been reached yet. It checks the current state of the world to see if the goal is already met. If not, it starts making a plan. This plan involves a sequence of actions based on the current state and what the AI can do. If the plan looks good, the AI starts with the first action. After each action, it looks at how the world has changed to make sure it’s getting closer to the goal.

The results of the actions are recorded in the AI’s diary, including what was done, the new state of the world, and whether the goal has been met. If the goal isn’t reached yet, the AI might change the plan or pick a new action to try. This loop continues until the goal is met or it seems impossible under the current conditions.

This methodical approach helps PhysarAI make steady progress towards goals but also shows its limitations in flexibility and adaptability in less controlled environments.