Comparison of two strategies of path planning for underwater robot navigation under uncertainty

This paper considers path planning for underwater robot in navigation tasks. The main challenge is how to deal with uncertainties in the underwater environment such as motion model error and sensing error. To overcome this challenge, two high level control methods have been presented and compared, which are based on the Model Predictive Control (MPC) strategy and the Partially Observable Markov Decision Process (POMDP) model, respectively. Navigation time, collision frequency, energy consumption and accuracy in localization are used as the assessment criteria for the two methods. It is shown that the MPC-based method is more efficient for our application scenarios while the POMDP-based method can provide more robust solutions.

I. Introduction

Underwater robot is a great application for robotics because working underwater is both dangerous and difficult for humans. Bridge piles cleaning is an important task for bridge inspection, maintenance, and rehabilitation. Fig. 1 shows the image of a bridge with piles covered by some marine growth. Carrying out high pressure underwater cleaning of piles can be a dangerous and exhaustive task for human and hence making use of underwater robot equipped with a water blasting gun can be beneficial and has large application potential. A basic requirement for such an underwater robot is that it must be able to move from its initial position to the destination (the pile that needs cleaning) without collision. Since there are a lot of uncertainties involved, such as motion error due to water current, the sensing errors, etc., the problem requires solving is path planning of underwater robot under uncertainties.