Thesis Work - Self-Adaptive Digital Twin for Robot Manipulation Through Continuous Real-Sim Discrepancy Learning

At ABB, we help industries outrun - leaner and cleaner. Here, progress is an expectation - for you, your team, and the world. As a global market leader, we’ll give you what you need to make it happen. It won’t always be easy, growing takes grit. But at ABB, you’ll never run alone. Run what runs the world.

This role sits within ABB's Robotics business, a leading global robotics company. We're entering an exciting new chapter as we’ve announced the plan for SoftBank Group to acquire ABB Robotics. SoftBank is a globally recognized technology group and investor/operator focused on AI, robotics, and next-generation computing.  By joining us now, you’ll be part of a pioneering team shaping the future of robotics—working alongside world-class experts in a fast-moving, innovation-driven environment.

This Position reports to:

R&D Center Lead

 

Your role and responsibilities

Digital Twins are increasingly used in industrial robotics for simulation, virtual commissioning, and robot learning. However, differences between simulated and physical systems can emerge due to uncertain or changing robot, task, and environment characteristics. These Real–Sim discrepancies can reduce the accuracy and reliability of Digital Twins and limit their use for Physical AI and Sim-to-Real applications.

This thesis investigates how information from a physical robotic system can be used to evaluate, adapt, and improve its Digital Twin, and how the improved twin can in turn support better prediction, validation, or learning for the physical system. The goal is to explore methods and architectures for establishing a bidirectional Real–Sim–Real interaction, where the Digital Twin can evolve based on physical experience rather than remaining a static model.

 

The student will investigate different approaches for identifying and reducing the Reality Gap. Possible directions include physics-based model adaptation, system identification, data-driven or learning-based methods, state estimation, hybrid physics–learning approaches, or combinations thereof. Architectures such as Shadow and Predictive Twins may also be explored, but the student is encouraged to investigate and compare alternative solutions.

 

The methodology should be generally applicable to industrial robotic systems. A dual-arm robotic system may be used as an experimental demonstrator.

Details:

• Period: January to July 2027
• Number of credits: 30 ECTS/högskolepoäng (hp)
• Number of students: 1
• Location: on-site, Västerås

The main tasks are:

• Literature review: Investigate the state of the art in bidirectional and adaptive Digital Twins, Real–Sim adaptation, simulation fidelity, system identification, and learning-based approaches for robotics.
• Problem formulation: Identify relevant sources of Real–Sim discrepancy and determine which robot, task, and environment characteristics are most important to observe and adapt.
• Method investigation: Explore and compare suitable approaches for using physical robot experience to evaluate and improve the Digital Twin. This may include physics-based, data-driven, learning-based, or hybrid approaches.
• Digital Twin framework: Design and implement a suitable bidirectional Real–Sim–Real framework based on the findings of the investigation.
• Evaluation: Define appropriate metrics for Digital Twin fidelity and evaluate the selected approach under controlled variations in the robot, task, or environment.
• Validation: Demonstrate and assess the developed framework on an industrial robotic manipulation scenario, potentially using a dual-arm robotic system.

The expected outcome is an investigation and experimental evaluation of methods for adaptive bidirectional Digital Twins, together with a demonstrated framework showing how physical robot experience can be used to improve Digital Twin fidelity and how the improved twin can support subsequent robot operation or Physical-AI applications.

The work can provide a foundation for future research in areas such as predictive simulation, synthetic data generation, robot learning, Sim-to-Real transfer, predictive validation, and continual learning.

Qualifications for the role

• Master students interested in robotics
• Python, C++, and ROS 2
• Robotics fundamentals (kinematics, dynamics, control)
• Experience with robot simulation tools is an advantage
• Basic knowledge of system identification, and machine learning is beneficial
• Linux

More about us

Recruiting Manager LiWei Qi, +46 73 021 2309, Supervisors: Aftab Ahmad, [email protected] Jonas Larsson [email protected] will answer your questions.

 

Positions are filled continuously. Please apply with your CV, academic transcripts, and a cover letter in English.

 

We look forward to receiving your application!

A Future Opportunity
Please note that this position is part of our talent pipeline and not an active job opening at this time. By applying, you express your interest in future career opportunities with ABB.

We value people from different backgrounds. Could this be your story? Apply today or visit www.abb.com to learn more about us and see the impact of our work across the globe.