Master's thesis: Resilient Vision-Language-Action Models for Real-World Collaborative Robots

Vision-Language-Action models connect visual input and language instructions to robot movements, but distributed deployment can introduce delays and safety risks. This thesis at RISE in Kista, in collaboration with KTH and within the Digital Futures EdgeWise project, explores how these models can reliably control collaborative robots despite limited connectivity and changing network conditions. The aim is to combine high-level cloud-based reasoning with responsive, safe motor control at the edge.

The assignment

• Study research on VLA models, imitation learning and distributed cloud-edge robotics, and become familiar with RISE's existing multi-agent prototypes.
• Gather and prepare robot demonstration data combining multiple camera views, robot state measurements and language instructions.
• Implement and optimise inference and control pipelines on edge hardware connected to physical robots.
• Evaluate communication, latency and buffering strategies under fluctuating connectivity to reduce interruptions in robot motion.
• Develop and test supervisory mechanisms that identify unsafe actions or execution failures and support partial task recovery.
• Run physical experiments measuring end-to-end delay, motion smoothness and success rates for multi-step manipulation tasks.
• Document the methods and results in a thesis report, and demonstrate the system at RISE and the partner university. The focus can be adapted to the student's interests or divided into complementary tracks for two students.

Who we are looking for

• One or two master's students studying Computer Science, Robotics, Machine Learning, Embedded Systems or a related engineering subject.
• Strong Python programming skills and practical experience with a modern deep learning framework such as PyTorch.
• Theoretical knowledge and practical coursework in machine learning, computer vision and/or robotics.
• Experience using Linux, Git and containerised or modular software architectures.
• Motivation to carry out laboratory experiments with physical robots and good spoken and written English.
• Additional useful experience includes robot learning tools, collaborative or industrial robots, motion control, kinematics, embedded model acceleration, multi-camera vision, multi-agent frameworks, network protocols or cloud-edge computing.

What we offer

• Supervision from Dr. Fehmi Ben Abdesslem, Dr. Mariano Scazzariello and Dr. Joakim Eriksson at RISE, coordinated with the university examiner.
• Access to advanced edge AI hardware, robotic manipulators, sensors, compute clusters and laboratory facilities at RISE and partner environments.
• A modular research scope with room to select a focus matching the student's background and interests.
• Financial compensation after approved completion: SEK 39,990 for an individual student, or SEK 30,000 per student when two students work together.

Practical details

30 ECTS, equivalent to approximately 20 weeks of full-time study. Flexible start by agreement, normally during the spring semester. Based in Kista, Stockholm, with remote study combined with on-site robotics experiments, hardware integration and testing. Suitable for one student or a pair working on complementary parts of the project. Apply with a short motivation letter outlining your background, interests and preferred direction, a CV and a recent academic transcript. Applications are reviewed on an ongoing basis until the position is filled; the final application date is 31 December 2026. Questions: Dr. Fehmi Ben Abdesslem, fehmi.ben.abdesslem@ri.se, or Dr. Mariano Scazzariello, mariano.scazzariello@ri.se.