Distributed AI Agent Architectures for 6G Robotic Control
| Reference number | |
| Coordinator | RISE Research Institutes of Sweden AB - RISE AB - Digitala System |
| Funding from Vinnova | SEK 99 946 |
| Project duration | February 2026 - September 2026 |
| Status | Completed |
| Venture | 6G - Competence supply |
| Call | 6G - Supervision of degree work |
Important results from the project
Yes, the goals were met. We successfully developed and evaluated a distributed AI architecture that decoupled heavy semantic planning (in the cloud) from local motor control (on an edge device) for a LeRobot SO-101 arm. The planner achieved 92% accuracy, and the robot executed physical tasks with up to 82% success. We also demonstrated that three cameras were crucial (increasing success from 18% to 75%) and mapped how network latency impacts the smoothness of robotic movements.
Expected long term effects
Long-term, the project demonstrates how advanced AI can control robots without expensive, energy-intensive onboard hardware by utilizing the Cloud-Edge continuum. This lowers the cost and climate footprint of future industrial robots. Our insights into network latency lay the groundwork for robust control via future 5G/6G networks. Furthermore, by open-sourcing our collected dataset, we contribute to accelerating Swedish and international research in robot learning.
Approach and implementation
The project was conducted as a master´s thesis by Jacopo Veronese at RISE. We built a software architecture that distributed the workload: a cloud-based "Planner Agent" (Gemma) created action plans, while a local "Action Agent" (Jetson edge node) controlled a LeRobot SO-101 arm. Data was collected via human teleoperation. The system was then evaluated through physical manipulation tasks (pick-and-place, object rotation), measuring task success, latency, and the impact of network delays.