An Approach to Task Representation Based on Object Features and Affordances

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Service robots require reliable task execution, human learning capabilities, and plan explainability.
• Existing methods for robot knowledge representation and extraction are often inadequate.

Purpose of the Study:

• To introduce a knowledge representation scheme for enhanced skill generalization and explainability in multi-purpose service robots.
• To develop techniques for extracting robot knowledge from raw data without requiring prior object information or 3D models.

Main Methods:

• Developed a novel knowledge representation scheme for robot scene understanding and task execution.
• Implemented techniques for extracting this knowledge from raw sensory data.
• Created a modular system with a computer vision system and a task reasoning module.

Main Results:

• The proposed knowledge representation facilitates skill generalization and explainability.
• The system successfully learned from a few demonstrations for tasks like item hanging and stacking.
• The approach does not require prior object knowledge or 3D models.

Conclusions:

• The novel knowledge representation scheme enhances robot reliability, learning, and human interaction.
• The modular architecture allows for easy integration of new recognition and reasoning routines.
• This research advances the development of more capable and understandable service robots.