Task Tree Retrieval for Robotic Cooking

TL;DR

This paper introduces FOON, a structured knowledge representation for robotic cooking, utilizing algorithms that assign weights based on manipulation success rates to improve task completion in kitchen robotics.

Contribution

It presents FOON, a novel knowledge structure derived from human manipulations, and three algorithms to enhance robotic cooking efficiency and success.

Findings

FOON effectively encodes kitchen tasks from videos.

Weighted algorithms reduce failure rates in robotic cooking.

The approach improves task success in robotic kitchen scenarios.

Abstract

Robotics is used to foster creativity. Humans can perform jobs in their unique manner, depending on the circumstances. This situation applies to food cooking. Robotic technology in the kitchen can speed up the process and reduce its workload. However, the potential of robotics in the kitchen is still unrealized. In this essay, the idea of FOON, a structural knowledge representation built on insights from human manipulations, is introduced. To reduce the failure rate and ensure that the task is effectively completed, three different algorithms have been implemented where weighted values have been assigned to the manipulations depending on the success rates of motion. This knowledge representation was created using videos of open-sourced recipes