Reinforcement Learning Driven Adaptive Sensing Strategy for Battery Aware IoT Devices

Battery efficiency is a fundamental challenge in the deployment of large-scale Internet of Things (IoT) systems, particularly in energy-constrained environments. Conventional sensing strategies, such as fixed-rate and duty-cycled approaches, often fail to adapt to dynamic power availability and contextual relevance of data, leading to rapid energy depletion or suboptimal sensing. To overcome this drawback of artificial intelligence, this paper presents RAS-BIoTNet, a new reinforcement learning-controlled adaptive sensing device to autonomously control sensor activities according to the realtime energy estimates, priorities of sensing activity, and operational conditions. The system uses Graph-Temporal Convolutional Networks and Bayesian LSTM to predict energy with great accuracy and Hierarchical Actor-Critic model to formulate optimal sensing policies in the event of multi-objective reward structures. The experimental findings indicate that RAS-BIoTNet is a better choice to use than the baseline strategies, as it increases battery lifetime by 27%, minimum task delay by 38.5%, and sensing accuracy by more than 96.3%, even when the situation is energy-critical. The model suggested will guarantee improved uptime of the device, enhance the quality of data and its responsiveness in real-time. This study opens the doors to smart sensing of the next-generation IoT ecosystem, in which the resource constraints play a vital role. The ability to adapt to the changing surroundings is the highlight of the system, where it can be widely applicable to smart cities, environmental monitoring, wearables in healthcare, and industrial IoT.