AutoGPT+P Affordance-based Task Planning with Large Language Models

Overview

This research paper proposes AutoGPT + P, a system that allows users to interact with robots using natural language.

AutoGPT+P uses affordances, which represent the actions that an object or environment can offer to an agent, to dynamically capture the initial state of a task planning problem.

This approach enables the system to handle planning with incomplete information, such as tasks with missing objects, by exploring the scene, suggesting alternatives, or providing a partial plan.

The system has been evaluated in simulations and tested on a humanoid robot, demonstrating its potential for real-world application.

The paper also compares AutoGPT+P to other state-of-the-art planning systems that use large language models (LLMs), showing its advantages in terms of accuracy and planning success rate.