Background
Healthcare and smart manufacturing are facing pressures from population aging, labor shortages, and rising quality requirements at the same time. Wards, long-term care facilities, and factory production lines are progressively introducing humanoid robots to assist with daily tasks. Yet, putting robots into actual use still requires solving three practical problems. First, conventional fingerprint-based indoor localization relies on manual surveying, and the survey has to be redone whenever the environment changes. If only WiFi, 5G, or BLE is used, the system is prone to interference and coverage blind spots, so stable localization is difficult to maintain in complex venues. Second, robots in clinical settings handle sensitive data such as patient records, movement trajectories, and medication histories, but most current systems lack institutional cybersecurity and privacy management, making them difficult to align with Taiwan's Personal Data Protection Act or the EU AI Act. Third, localization, autonomous navigation, and task planning all depend on AI decisions. If the behavior goes out of control, it could affect personnel safety, yet Taiwan currently does not have a dedicated management framework for robotic AI systems. To address these issues, this sub-project combines three things, AI-adaptive localization, the ISO international certification system, and the cloud-native Agent Operation service, into a shared infrastructure that other sub-projects in the Center can directly rely on when adopting robots.
Research Objectives
Complete the AI-adaptive wireless localization system and verify it in real fields. The two-phase architecture includes "Rapid Field Deployment" and "Sustainable Optimization Closed-Loop," targeting initial-model generation within 10 minutes, localization accuracy within 3 m, inference latency within 10 ms, and a model that updates itself as the robot is used.
Complete the gap analysis, system establishment, and submission for four ISO international certifications: ISO 42001 (AI Management), ISO 27001 (Information Security), ISO 27701 (Privacy), and ISO 17025 (Laboratory Measurement). The goal is to become the first robotics R&D laboratory in Taiwan to hold all four certifications.
Complete the unified Agent Operation service platform that provides version control, Prompt and Model Registries, and automated CI/CD pipelines for AI Agents across the Center's sub-projects, and use Kubernetes for cross-cluster deployment.
Complete the multi-robot intelligent orchestration architecture. Build Coordinator and Robot Agents on top of the Agent-to-Agent (A2A) protocol, integrate heterogeneous robots, including the medication robots in Sub-project 4 and the sensing platforms in Sub-projects 5 and 7, and verify the closed-loop flow from task parsing to dispatch and reporting.
Complete the AI-RAN benchmarking framework and present the results internationally. Establish unified metrics for comparing DNN, CNN, and Transformer models, demonstrate them at MWC 2026 and the O-RAN Alliance Virtual Exhibition, and continue to work with iCAIR, Chunghwa Telecom Research Institute, and Chin Chun Enterprise (Thailand).
Methods
Multi-modal signal fusion for localization. Use WiFi RSSI, 5G channel state, BLE signal strength, and magnetic-field vectors together so that the system is not affected when any single signal is interfered with or unavailable.
Rapid field deployment. Robots equipped with multi-modal receivers and an IMU use Pedestrian Dead Reckoning (PDR) to align signals to field coordinates automatically. The Central Station back-end handles grid management, feature analysis, and model training, so two robots can finish the initial setup of a 1,000-ping site in a single day.
Sustainable optimization closed-loop. Semi-supervised learning with confidence-weighted feedback automatically routes signals collected during routine missions back into model retraining and redeployment, forming a "collect, train, deploy" loop that does not require manual labeling.
AI-RAN benchmarking and data sovereignty audit. Run edge-to-cloud closed-loop tests using unified accuracy, latency, and jitter metrics, and introduce a Data Sovereignty Technical Audit (DSTA) aligned with NIST IR 8425 to ensure non-red supply chain compliance and end-to-end encrypted transmission.
Cloud-native and multi-agent orchestration. The Agent Operation platform is built on Docker, Kubernetes, and AgentOps/MLOps. On top of that, a Coordinator Agent is implemented using the A2A protocol with Google ADK, together with LLM-driven semantic navigation and the AG-UI interface, to coordinate multiple Robot Agents across cross-robot tasks.
ISO four-certification implementation. Lifecycle risk assessment, Privacy Impact Assessment (PIA), measurement uncertainty analysis, and AI bias and fairness controls are applied to fulfill the requirements of ISO 27001, 27701, 17025, and 42001 respectively. Lab 611 is upgraded into the Open Communication Architecture Testing Laboratory (OCATL) as the empirical site.
Innovation
Sustainably optimized AI-adaptive localization closed-loop. During task execution, robots continuously collect WiFi, 5G, BLE, and magnetic-field signals. With semi-supervised learning and automatic feedback deployment, the localization model improves over time, achieving the "improving-with-use" effect without manual labeling after deployment.
Deployment-oriented AI-RAN benchmarking framework. Designed for 6G AI-RAN indoor localization, it compares DNN, CNN, and Transformer models with unified metrics including accuracy, latency, and jitter, and reaches sub-10 ms inference latency and sub-3 m localization accuracy in edge-to-cloud closed-loop verification.
Data Sovereignty Technical Audit (DSTA) architecture. Aligned with NIST IR 8425, it ensures non-red supply chain compliance and end-to-end encrypted transmission.
Cross-sub-project Agent Operation platform. A Coordinator Agent built on the A2A protocol encapsulates heterogeneous robots as independent Robot Agents, delivering standardized orchestration across multiple robots, with shared CI/CD pipelines and Kubernetes deployment available to all sub-projects in the Center.
Expected Outcomes
Technical breakthrough. The AI-adaptive wireless localization system is completed. Robots can produce an initial localization model within 10 minutes, with localization accuracy better than 3 m and inference latency below 10 ms, suitable for indoor navigation, smart-field management, and task scheduling.
International certification milestone. We are pursuing ISO 42001, ISO 27001, ISO 27701, and ISO 17025, aiming to be the first robotics R&D laboratory in Taiwan that holds all four certifications.
Industry-academia engagement. Industry-academia collaboration and technology-transfer revenue total NT$6.51M, NSTC funding contributes NT$3.85M, for a combined NT$10.36M. Partnerships have been established with Chunghwa Telecom Research Institute, Northwestern University iCAIR, and Chin Chun Enterprise (Thailand).
Academic output. 1 international journal paper and 5 international conference papers have been published. 7 industry engineers have been trained, and several graduate-level researchers have been cultivated.
International engagement. The team actively participates in the technical discussions of O-RAN ALLIANCE and AI-RAN Alliance, and exhibits at the O-RAN Alliance Virtual Exhibition to raise Taiwan's visibility in open communications architecture.