EdgeWise: Secure AI Models for Humanoid Robots

The Need for Secure and Resilient VLA Systems

Humanoid robots powered by Vision-Language-Action models have the potential to transform areas such as healthcare, construction, search and rescue, and other labour-intensive or hazardous fields. However, today’s VLA architectures rely heavily on cloud-based reasoning models, which introduce latency and require stable connectivity. This makes them less suitable for many real-world environments, such as disaster zones, underground mines or remote healthcare settings.

VLA systems also often generate action representations that are difficult to interpret, review or verify, creating significant safety and security risks. Recent studies show that robots controlled by language models can be manipulated into carrying out unsafe actions, highlighting the need for greater transparency and formal safety guarantees.

EdgeWise addresses these limitations by combining edge computing, efficient model-sharing mechanisms and formal verification methods to enable secure, resilient robotic systems with low latency.

Project Objectives

The main objectives are to:

• Design a new three-tier VLA architecture spanning on-device, edge and cloud deployment to reduce end-to-end control-loop latency and improve resilience under unstable network conditions.
• Enable efficient use of edge models across multiple concurrent users by compressing fine-tuned models into lightweight adapters that share a common base model, significantly improving resource utilisation.
• Introduce verifiable and interpretable execution planning, where robots generate structured code from natural-language instructions and apply formal verification methods to ensure correctness and safety before execution.

Cross-Disciplinary Collaboration

EdgeWise is a collaboration between RISE and KTH, bringing together expertise in:

• distributed systems and edge-cloud orchestration
• machine learning systems and large-model deployment
• formal verification and secure software systems
• robotics and experimental testbeds

The project brings together research in systems, AI model optimisation, networking technologies such as 5G/6G and formal methods. The experimental platform includes robotic systems connected to private 5G infrastructure and research data centre resources, enabling controlled evaluation of latency, resilience and safety mechanisms.