ACAMR: AI-Enabled Communication Algorithm in ROS to Improve Mobile Robot Connectivity in Internet of Robotic Things for AMVs

Abstract Autonomous Marine Vehicles (AMVs) are crucial in various maritime applications, including ocean exploration, environmental monitoring and offshore infrastructure inspections. Existing approaches to robot navigation such as reinforcement learning and supervised learning, often face challenges in dynamic environments due to high training overhead, poor generalization, and a lack of real-time adaptability. Federated learning techniques partially address privacy concerns but still incur significant communication costs and latency. Centralized architectures further suffer from bottlenecks and single-point failures, making them less suitable for mission-critical tasks. The framework integrates sensor fusion techniques, predictive analytics, and reinforcement learning-based FTC algorithms to enable AMVs to recover from propulsion, sensor, and communication failures. Furthermore, the incorporation of Internet of Robotic Things (IoRT) allows distributed fault monitoring, cooperative decision-making, and secure cloud-based diagnostics, enhancing the resilience of AMV fleets. The goal is to take advantage of both the ROS’s strengths in robotic control and coordination with the IoT’s strengths in connecting devices and exchanging data efficiently. In the following work, an integrated framework is proposed that allows ROS-enabled robots to communicate and work together using IoT protocols and the infrastructure of AMVs. Robots are able to learn from their environments, adapt to new situations, and complete collaborative tasks more effectively and independently. SLAM, Rviz and ROS2 have been discussed and validation of the proposed work over state of art approaches is shown for navigation of mobile robots in Gazebo software. Experimental results demonstrate a 34% reduction in navigation time, 29% improvement in distance efficiency, and 42% increase in task success rate compared to baseline learning models.