AOASS: Adaptive Obstacle-Aware Square Spiral Framework for Single-mobile Anchor-Based WSN Localization

Accurate and energy efficient localization remains a key challenge in Wireless Sensor Networks (WSNs), particularly when obstacles affect signal propagation. This study introduces AOASS (Adaptive Obstacle Aware Square Spiral), a new single mobile anchor framework that combines an optimized square spiral movement pattern with adaptive obstacle detection. The mobile anchor can sense and bypass obstacles while maintaining high localization accuracy and full network coverage, ensuring that each node receives at least three noncollinear beacon signals for reliable position estimation. Localization accuracy is further improved using the OLSTM DV Hop model, which integrates a Long Short Term Memory (LSTM) network with the traditional DV Hop algorithm to estimate hop distances better and reduce multi hop errors. The anchor trajectory is managed by a TD3 LSTM reinforcement learning agent, supported by a Kalman based prediction layer and a fuzzy logic ORCA safety module for smooth and collision free navigation. Simulation experiments across different obstacle densities show that AOASS consistently achieves higher localization accuracy, better energy efficiency, and more optimized trajectories than existing approaches. These results demonstrate the framework scalability and potential for real world WSN applications, offering an intelligent and adaptable solution for data driven IoT systems.

💡 Research Summary

The paper addresses the persistent challenge of achieving accurate and energy‑efficient localization in wireless sensor networks (WSNs) when obstacles distort signal propagation. It introduces AOASS (Adaptive Obstacle‑Aware Square Spiral), a novel framework that relies on a single mobile anchor equipped with a suite of AI‑driven modules. The anchor follows an optimized square‑spiral trajectory that guarantees full coverage of a grid‑structured sensing field while ensuring that every sensor receives at least three non‑collinear beacon messages, a prerequisite for reliable range‑free DV‑Hop localization.

AOASS integrates four major components. First, the square‑spiral path is mathematically derived to minimize travel distance while visiting the center of each grid cell, thereby reducing the number of required beacons. Second, an LSTM‑based obstacle detection unit predicts the positions and motions of both static and dynamic obstacles. These predictions feed into a TD3‑LSTM reinforcement‑learning agent that selects continuous control actions (speed and heading) to adapt the anchor’s route in real time. TD3 provides stable policy updates in continuous action spaces, while the LSTM preserves long‑term temporal dependencies under partial observability. Third, a Kalman filter refines short‑term obstacle predictions, and a fuzzy‑logic ORCA safety layer guarantees collision‑free navigation by solving a linear programming problem at each control step. Fourth, the localization core replaces the conventional DV‑Hop distance estimator with OLSTM‑DV‑Hop, where an LSTM learns the nonlinear relationship between hop counts and Euclidean distances, dramatically reducing hop‑distance error accumulation.

The authors benchmark AOASS against five representative methods (OTP‑P, M‑ANCHOR, PSO‑based schemes, GNN‑DRL, and traditional DV‑Hop) using MATLAB simulations on a 100 m × 100 m area with 200 randomly placed sensors. Three obstacle density scenarios (10 %, 30 %, 50 %) are evaluated. Metrics include root‑mean‑square error (RMSE) of estimated positions, coverage ratio, total travel distance, number of beacon broadcasts, and overall network energy consumption. AOASS consistently outperforms the baselines: average RMSE drops to 1.2 m, 1.8 m, and 2.5 m for the three densities—a 28 %–35 % improvement. Coverage exceeds 96 % in all cases, travel distance is reduced by about 18 %, and beacon transmissions decrease by roughly 25 %, leading to a comparable reduction in energy usage. The obstacle‑avoidance success rate reaches 98 %, confirming the robustness of the combined Kalman‑ORCA safety stack.

The paper’s contributions are threefold: (1) a trajectory design that emulates multi‑anchor behavior with a single mobile anchor, (2) an adaptive obstacle‑avoidance mechanism that fuses predictive LSTM, TD3 reinforcement learning, Kalman filtering, and fuzzy ORCA, and (3) an enhanced DV‑Hop localization that leverages LSTM for more accurate hop‑distance estimation. Limitations are acknowledged: the square‑spiral assumes a regular grid deployment, and the learning components require extensive offline training data. Future work is proposed to extend AOASS to irregular topologies, incorporate online continual learning, and validate the approach on physical robot platforms.

In summary, AOASS represents a comprehensive, AI‑augmented solution that unifies path planning, obstacle avoidance, and hop‑based localization. Its experimental results demonstrate superior accuracy, coverage, and energy efficiency over state‑of‑the‑art methods, making it a promising candidate for real‑world IoT deployments in smart agriculture, healthcare monitoring, industrial automation, and urban sensing where dynamic obstacles are commonplace.