Convolutional Long Short-Term Memory Predictor for Collaborative Remotely Operated Vehicle Trajectory Tracking in a Leader–Follower Formation Subject to Communication and Sensor Latency in the Presence of External Disturbances

Abstract

Nowadays, collaborative operations between Remotely Operated Vehicles (ROVs) face considerable challenges, particularly in leader–follower schemes. The underwater environment imposes limitations on acoustic modems, leading to reduced transmission speeds and increased latency in ROV position and speed transmission. This complicates effective communication between the ROVs. Traditional methods, such as Recursive Least Squares (RLS) predictors and the Kalman filter, have been employed to address these issues. However, these approaches have limitations in handling non-linear patterns and disturbances in underwater environments. This paper introduces a Convolutional Long Short-Term Memory (ConvLSTM) predictor designed to enhance communication and trajectory tracking between ROVs in a leader–follower scheme. The proposed ConvLSTM aims to address the shortcomings of previous methods by adapting effectively to varying conditions, including ocean currents, communication delays, and signal interruptions. Simulations were conducted to evaluate ConvLSTM’s performance and compare it with other advanced predictors, such as Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM), under different conditions. The results demonstrated that ConvLSTM achieved a 13.9% improvement in trajectory tracking, surpassing other predictors in scenarios that replicate real underwater conditions and multi-vehicle communication. These results highlight ConvLSTM’s potential to significantly enhance the performance and stability of collaborative ROV operations in dynamic underwater environments.