Cobots learning by demonstration
Start date: July 2022
Expected end date: December 2025
Robots in industrial applications struggle with limited flexibility, high programming complexity, and workforce skill shortages. Traditional programming methods are inefficient for dynamic environments, making automation inaccessible to non-experts. To address this, Learning from Demonstration (LfD) allows robots to learn tasks by observing human actions, minimizing the need for manual coding. However, existing LfD techniques demand large datasets, intensive computation, and struggle with generalization.
This research focuses on a Minimal Key-Point Learning Framework for robotic skill transfer from videos. Using vision-based tracking, the system extracts essential motion data, significantly reducing dataset requirements. Open-source and pre-trained components minimize computational burden while ensuring high accuracy in action recognition and segmentation. The framework decomposes tasks into smaller, executable sub-tasks, enabling faster and more adaptable robot learning.
By leveraging minimal data and lightweight processing, this approach makes robot training more efficient and accessible, bridging the gap between human demonstration and robotic execution for real-world automation.
Expected Outcomes
This research will contribute to:
- A Minimal Key-Point Learning Framework that simplifies robotic skill transfer from videos with reduced data and computation.
- Improved pose-tracking-based action recognition, optimizing accuracy with minimal input features.
- A refined task segmentation approach, enabling effective breakdown and execution of complex tasks.
- A scalable implementation strategy, making robot programming more intuitive for non-experts.
By addressing these challenges, this study will contribute to the development of more adaptable and efficient robotic learning systems, ensuring seamless integration of human-robot collaboration in industrial environments.
Supervisory Team
- Principal Supervisor: Professor Mats Isaksson
- Associate Supervisors: Professor John McCormick
- External Supervisor: Dr Fouad Sukkar
Publications
- McCormick, John, Pyaraka, Jagannatha Chargee, Real Robot Dance Challenge: Exploring live and online dance challenge videos for robot movement, International Conference on Movement and Computing (MOCO 2024), Utrecht, Netherlands.
- McCormick, John, Pyaraka, Jagannatha Chargee, Choudray, Kartik, Track Back: A Human Robot Movement Installation Utilising Unity Digital Twin and Human Bio-mimicry, International Symposium on Electronic Art (ISEA 2024), Brisbane, Australia.
-
Kartik Choudhary, Jagannatha Charjee Pyaraka, Mats Isaksson, Development of an Autonomous Mobile Cobotic Platform for Targeted Herbicide Dispersion, Australasian Conference on Robotics and Automation (ACRA 2023). ISBN: 9780645565522
Associated Researchers