Project: Enhancing Real-World Imitation Learning with Reinforcement Learning
Description
This TU/e master project is setup in collaboration with a robotics start-up in Eindhoven.
Company Overview
TeleOperation Services is an innovative company based in Woensel-Noord, Eindhoven. Our cutting-edge AI-driven system empowers robotic arms to imitate tasks and perform them independently with human-like finesse and speed. Through our unique leasing model, we make advanced robotic technology accessible without significant investments. We are revolutionizing the deployment of robotic arms by leveraging teleoperation, allowing complex tasks to be automated without any programming.
Motivation
While our system achieves an impressive 99.5% success rate with simple applications using imitation learning, this rate drops to 95% for more complex actions which is too little for industry applications. Our ambition is to push this success rate towards 100%, while enhancing the model's ability to handle increasingly complex and unfamiliar situations. Achieving this will significantly advance the reliability and versatility of our robotic systems in real-world applications. Feel free to ask the company supervisor for videos of use-cases.
Challenge
We believe that integrating reinforcement learning with our existing imitation learning models can substantially improve performance over time. By refining the imitation learning architecture and parameters, and incorporating reinforcement learning techniques, we aim to optimize decisiveness and accuracy in the actions of the robotic arm, and to enhance the model's adaptability and learning capability to complex and ‘unknown’ scenarios.
Objectives
You will conduct research and development by investigating methods that combine imitation learning with reinforcement learning to boost model performance in complex tasks. This involves developing and integrating new algorithms or architectures into our existing AI framework to enhance robotic decision-making processes and performance. You will conduct real-world experiments on state-of-the-art robotic systems to validate improvements in reliability and success rates. Additionally, ensuring that the solutions are scalable and effective in increasingly complex environments will be a key aspect of the project.
Prerequisites
- Proficiency in programming languages such as Python and/or C/C++.
- Experience with machine learning frameworks (e.g., PyTorch, TensorFlow).
- Analytical mindset with a passion for robotics and AI innovation.
Do you want to apply or do you want more information?
Company Supervisor: Teun Jansen
Email: T.Jansen@TeleOperationServices.com
Phone: +31 6 48872523
Details
- Supervisor
-
Bram Grooten
- Secondary supervisor
-
Thiago Simão
- External location
- TeleOperation Services
- Interested?
- Get in contact