Tojoarisoa Rakotoaritina

In this study, we seek to determine which intrinsic rewards are most effective in different kinds of environments. We formulate three typical intrinsic rewards in a common information-theoretical basis: (1) \textbfnovelty as low probability of observation; (2) \textbfsurprise as low probability of state transition; and (3) \textbfempowerment as maximum mutual information between actions and subsequent states. We implemented the framework using a combination of Variational autoencoder-based Observation Density Model (VODM), and a Variational Forward Dynamics Model (VFDM). We tested this framework in three MiniGrid "Four Rooms" environments with sparse rewards under varying stationarity and subgoal conditions. Results showed that a combination of three intrinsic rewards enhances learning in the environment with hidden subgoals, such as key-and-door mechanics.

Wildfires are escalating in frequency and severity, particularly in high-risk regions such as Alberta, Canada, where traditional detection systems are becoming increasingly insufficient. Existing approaches often rely on centralized control or overlook key constraints, such as partial observability, terrain complexity, and communication limitations. To address this gap, we propose a fully decentralized multi-agent reinforcement learning (MARL) framework for wildfire detection using UAV swarms. Our method integrates real geographic data into a grid-based simulator and employs intrinsic-motivation-enhanced Independent Proximal Policy Optimization (IPPO), allowing each agent to learn independently and adaptively. This design is well-suited for large-scale, unstructured environments where centralized coordination is infeasible. Agents learn to balance exploration, fire detection, and risk mitigation through a hybrid reward scheme. Experimental results in simulation demonstrate the effectiveness of our method for early and reliable wildfire detection in large, remote landscapes. This work lays the foundation for scalable, robust, and communication-efficient UAV swarm systems for wildfire monitoring, with significant potential to reduce ecological, economic, and human costs.

Nowadays, within the concept of Internet of Things (IoT), smart homes, smart factory, intelligent transportation among others are infrastructure systems that connect our world to the Internet. However, wireless communications technology are considerably constrained by complicated structures, and lossy media in complex environments. Fundamental limitations on the transmission range have been treated to connect IoT devices in such Radio Frequency (RF) challenging environments. In order to extend the transmission range in complex environments, Magnetic Induction (MI) communication has been proved to be an efficient solution. In this paper, a Multilayered Metamaterial low profile antenna (MMLPA) using Magnetic Induction communication scheme is proposed for IoT applications. The system model of the MMLPA is analyzed. Then an MMLPA system is designed by using a circular loop antenna backed with isotropic metamaterial which is considered as a Defected Ground Structure (DGS) as well as with anisotropic metamaterial for the purpose of a dielectric uniaxial metamaterial. By using a full-wave finite-element method, the proposed analysis is supported with simulation results where good agreement is achieved compared to the measurement results after realizing four prototypes of the MMLPA antennas. The effect of the presence of metal in the vicinity of the transceivers is also analyzed.