Fair and Efficient Scheduling for Sensor Networks via Online Whittle Index Policy

We study the problem of intelligent polling in wireless sensor networks, where transmissions should occur only when updates are meaningful. To prioritise informative updates, we formulate the polling problem using the Age of Incorrect Information (AoII) metric and cast it as a restless multi-armed bandit problem. We prove that the AoII-based polling problem is indexable and admits a threshold policy structure. Building on this, we derive an online Whittle index approach that uses real-time state estimation to select nodes for polling. Crucially, our method does not assume knowledge of the system's transition dynamics. We further show that the optimal penalty threshold for this index policy can be dynamically learned, enabling adaptive and low-complexity operation in realistic environments. To demonstrate the practical relevance of our approach, we apply the technique to a Wake Up Radio (WUR) system and show that it reduces packet transmissions by up to 70% while maintaining accurate reconstruction with low Root Mean Squared Error (RMSE) an aspect not considered by other polling techniques. These results validated on both synthetic and real sensor data highlight the effectiveness of our method for goal-oriented, energy-aware communication.