A Case Study: Modeling of A Passive Flexible Link on A Floating Platform for Intervention Tasks

Modular Transfer Learning with Transition Mismatch Compensation for Excessive Disturbance Rejection

“…In each iteration, more training samples are generated following the previous procedures, but the disturbances used by GCP now come from the predicted disturbance parameters of ODI μ, rather than the true disturbance parameters μ used in the system dynamics. Then, ODI is trained again through combining the mismatched training samples with previously gathered ones according to (6). After a small number of iterations, the combined system, GCP-ODI, achieves close performance with GCP that is fed with the true disturbance parameters μ.…”

“…Owing to this decay, as well as the considerable size and thrust capabilities of underwater robotic systems, the strength and changes of ocean waves are often neglected in robot motion planning and control in deep water applications [4]. In field applications with low operational depths and turbulent wave climates, like bridge pile inspection [5] and sea-ice algae characterization in Antarctica [6], this assumption can quickly break down, since shallow water environments usually accommodate only small-size robots that have limited thrust capabilities, and the disturbances coming from the turbulent flows are time-varying and may frequently exceed robot's thrust capabilities (such wave forces are termed as excessive disturbances throughout this paper). As a result, increased wave forces inevitably hinder the stability and precision of robot motion control [7]- [9].…”