Tangram Collaboration Platform

The Tangram Collaboration Platform is a research project that combines AI assistance with a turn-based game interface to collaboratively build tangram puzzles. This platform integrates a chat-based interface with a dynamic game area and an external Python AI agent (communicating via WebSockets) to offer an engaging, interactive puzzle-building experience.

Table of Contents

• Overview
• Features
• AI Agents
• Setup and Installation
• Developer API

Overview

This platform enables users to collaborate with AI assistants in a turn-based tangram puzzle-solving experience. The platform encourages creative problem-solving and collaboration through an intuitive interface and advanced AI support and provides a platform for experimentation with creative AI agents with visual resources.

Features

The platform offers the following key features:

• Turn-Based Gameplay: Alternating moves between human players and AI agents.
• Real-Time Chat: A built-in chatbox for strategy planning, explanations, and idea sharing.
• Game State: A game board and a “drawer” for unplaced pieces, visible to users and AI agents with visual capabilities as well as a coordinate system.
• Modular Architecture: Separation between the game and AI agents to allow for experimentation with different agents, complete with error warning from the platform.

AI Agents

The current implementation includes two AI agents:

• SimpleGPTAgent: Powered by GPT-4, this agent reviews its own plays to self-correct and optimize its strategies.
• RelationalAgentGPT: Also based on GPT-4, this agent leverages spatial descriptions relative to the current board state to generate informed moves.

Setup and Installation

For instructions on running the current version of the game, please refer to the provided setup guide.

A demo of an older version of the platform is available, offering a glimpse of the interactive experience. Please note however that this version is outdated and does not reflect the current state of the project:

Video Showcases

Iterative Simple Agent

Relational Agent

missing video

Developer API

Developers looking to create new AI agents can find detailed documentation in the API.md file. This document outlines the developer API, including communication protocols and integration guidelines.