Master's thesis; Multi-Agent AI Architectures for Real-World Robotics

Background

As monolithic AI models reach their limits in complex, dynamic environments, the future of robotic intelligence lies in creating teams of specialised agents that can collaborate to solve problems. This approach mirrors how human teams work, allowing for more robust, safe, and nuanced behaviours. This thesis project will explore this frontier by designing and building a multi-agent Vision-Language Agent (VLA) architecture for a real-world robot.

Project Description

This Master's thesis will focus on creating a "collaborative mind" for a single robot. Instead of relying on one large AI, the project will develop a system where multiple, specialised AI agents work together. Drawing inspiration and tools from state-of-the-art open-source frameworks like LeRobot, the goal is to create a decentralised intelligence where agents can take on distinct roles, such as:

• A "Planner" Agent that sets high-level strategic goals.
• A "Controller" Agent that executes precise, low-level actions.
• A "Safety" Agent that acts as a supervisor with the power to override commands to prevent hazards.

The core scientific challenge is to design the communication and negotiation protocols that allow these agents to function as a cohesive unit and to benchmark this paradigm against traditional, single-agent architectures.

Key Responsibilities

• Design and implement a multi-agent VLA framework for a robotic platform.

• Fine-tune pre-trained VLA models for specialised roles using Imitation and Reinforcement Learning techniques.

• Utilise the lerobot library to deploy and test the architecture on a physical robot.

• Rigorously analyse the trade-offs in performance, safety, and robustness compared to a single-agent system.

• Collaborate with senior researchers at RISE.

Qualifications

• Ongoing Master’s degree in Computer Science, AI, Robotics, or a related field.

• A strong background in AI, deep learning, and computer vision.

• Proficiency in Python and experience with ML frameworks like PyTorch or TensorFlow.

• A keen interest in robotics, autonomous systems, and AI agent architectures.

• Experience with Reinforcement/Imitation Learning or the Hugging Face ecosystem is a plus, but not a prerequisite.

• A creative, analytical, and collaborative mindset.

What we offer

• A chance to work on a scientifically novel project that is defining the future of AI in robotics.

• Hands-on experience with state-of-the-art open-source robotics frameworks.

• Access to RISE’s advanced resources, unique datasets, and test environments.

• Opportunities for professional growth and networking, including potential PhD opportunities.

• A friendly and dynamic research environment with experienced supervisors.

Terms

• Scope: The thesis usually comprises 30 credits (hp/ECTS).

• Start Date: Spring 2026, or by agreement.

• Location: RISE, Kista, Stockholm, with the possibility for some remote work.

• Compensation: In line with RISE guidelines for strategically important projects, a compensation of 39,990 SEK is offered upon completion and approval of the 30-credit thesis.

• Please note that due to industrial confidentiality and non-disclosure agreements, this position is restricted to EU citizens only.

Welcome with your application!

Applications must be submitted through the Teamtailor recruitment portal.

Last day of application: 27th of October, 2025.

For more information, please contact:

• Dr Fehmi Ben Abdesslem, fehmi.ben.abdesslem@ri.se

• Dr Joakim Eriksson, joakim.eriksson@ri.se

• Miriana Passarotto, miriana.passarotto@ri.se