José Jaime da Cruz

Optimal control of ship unloaders using reinforcement learning

Abstract This paper describes the use of Reinforcement Learning (RL) to the computation of time-optimal anti-swing control of a ship unloader. The unloading cycle has been divided into six subtasks and an optimization problem has been defined for each of them. A RL algorithm together with a multilayer perceptron neural network as a value function approximator have been used in the optimization. The results obtained are encouraging, since they reproduce a solution previously generated by using Optimal Control Theory.

Study of autonomous underwater vehicle wave disturbance rejection in the diving plane

Autonomous underwater vehicles operating near the sea surface are subject to wave disturbances in their motion, which can affect severely the data acquisition during the vehicle motion. Bathymetric maps, for example, may suffer a big loss in quality due to such kind of disturbances that can impose oscillations in the vehicle depth and attitude. This article describes a controller structure for compensating the wave disturbances in the vertical plane for a torpedo-like autonomous underwater vehicle, based on a modified version of the linear quadratic Gaussian with loop transfer recovery control combined with a wave filter. The proposed methodology uses additional to the classic linear quadratic Gaussian with loop transfer recovery design the measurement of non-controlled states, such as the pitch rate and heave velocity, in the control action. The wave disturbances were filtered away by a shaping filter fitted to the sea spectrum, and the sensor signals were integrated by an extended Kalman filter. The proposed control methodology was tested against the classical linear quadratic Gaussian with loop transfer recovery controller in a 6-degree-of-freedom non-linear autonomous underwater vehicle simulator, producing a better performance in most demanding conditions such high wave crests. The tests also revealed the impact of the hydroplane geometry on the controller performance, which should not be underestimated by the autopilot designer.

Complete offline tuning of the unscented Kalman filter

Abstract The unscented Kalman filter (UKF) is a widely used nonlinear Gaussian filter. It has the potential to deal with highly nonlinear dynamic systems, while displaying computational cost of the same order of magnitude as that of the extended Kalman filter (EKF). The quality of the estimates produced by the UKF is dependent on the tuning of both the parameters that govern the unscented transform (UT) and the two noise covariance matrices of the system model. In this paper, the tuning of the UKF is framed as an optimization problem. The tuning problem is solved by a new stochastic search algorithm and by a standard model-based optimizer. The filters tuned with the proposed algorithm and with the standard model-based optimizer are numerically tested against other nonlinear Gaussian filters, including two UKF tuned with state-of-the-art tuning strategies. One of these strategies relies on online tuning and the other on offline tuning.

Optimal power loss motion planning in legged robots

An iterative algorithm to minimize energy loss in kinematic chains is proposed. This algorithm is designed to low level of control where variables such as terminal states, runtime, and physical and electrical parameters of the movement are given by higher levels of control. An original complex problem of optimization is transformed into a simple quadratic programming problem subject to linear constraints by discretizing all dynamic system variables. The whole system is then converted into a recursive matrix equation that is solved iteratively. A proof of convergence is suggested. The performance of the algorithm is illustrated by using it in the motion planning of a quadruped robot.

Kinematics Analysis of a Quadruped Robot

Abstract This paper presents the kinematics model of a complex quadruped robot. Initially, the so-called position analysis problem is approached for both the inverse and the direct kinematics problems. All equations and geometric constraints of the system are presented. Finally a singularity analysis of closed-chain kinematics is done, pointing out which configurations should be used or avoided for a desired behavior of the robot. The singularity analysis was done by considering three different kinds of singularities: direct, inverse and combined.

A Tree-Climbing Robot Platform: Mechanical Concept, Control Software and Electronic Architectures

Robotic platforms can be used in countless applications and in the most varied branches of activities. Such results, presented over the last two decades, can be verified in Armada et al., 2003 and Virk, 2005. It deals specifically with robots provided with legs and with the capability, or ability, to climb vertical surfaces, while many other applications/solutions can be found. As an examples: robots to climb lower parts of bridges (Abderrahim et al., 1999), to crawl inside pipes for inspection purposes (Galvez et al., 2001), implemented to perform solder inspection works in nuclear plants (White et al., 1998), to climb metallic structures for inspection purposes (Armada et al., 1990). More examples can be found in marine industry applications, such as walking robots to check internal parts solders in ship hulls, climbing robots for parts solders (Santos et al., 2000) and (Armada et al., 2005), climbing robots for paint cleaning, underwater robots for ballast tank inspection, and underwater robots for hull cleaning (Santos et al., 1997a, 1997b). Robots were demonstrated to be the ideal option for many such applications due to the fact that the working environment is difficult to access or even hazardous or risky for human beings, such as exposure to hazardous substances or environments and risk conditions. Productivity increase and quality issues are also extremely relevant and are considered. However, besides the varied applications and areas mentioned above, there still remains a little explored area: environmental research. As in any other area, different applications or problems can be addressed or solved with the help of a robotics platform. As an example, one can mention the tasks: • Gathering of Botanical Specimens: gathering flower and plant specimen is fundamental for biodiversity studies. Several host species are found in high trees, and their collection is actually risky. Thus, this is an application where a robotics platform with tree climbing capability can be used to minimize risks for the researchers involved in this type of activity. • Gathering of vegetable material: collection of vegetable material is a fundamental activity for phytochemistry. Every pharmaceutical preparation that uses as raw material plant parts such as leaves, stems, roots, flowers, and seeds, with known O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m

Development of an On-Board Wave Estimation System Based on the Motions of a Moored FPSO: Preliminary Assessments From a Field Campaign

This paper addresses a wave inference system developed for on-site estimation of directional wave spectra from the measurements of the wave-frequency motions of moored vessels. As a preliminary evaluation of the system’s performance, the results obtained in a 9-month field campaign based on an FPSO operating in Brazil’s Campos Basin are compared to hindcast predictions for the same region displayed by NOAA. The estimation method is based on Bayesian inference algorithms, previously validated by means of numerical and small-scale experimental analysis. A 6-dof inertial measurement unit (IMU) is used for monitoring the motions of the platform. The algorithm also requires the Response Amplitude Operators (RAOs), which in turn depend on the loading conditions of the FPSO. Those are defined before each estimation is performed by means of readings of the tanks’ levels. The accuracy of the predictions also depends on the calibration of two parameters (the so-called hyperparameters of the Bayesian method) and those are specified with respect to the mean period of heave motion. The system is able to identify unimodal or bimodal (cross) spectra using a bimodality criterion combined to an iterative algorithm for separating each component of the global spectrum. As a result, besides the global energy matrix, the statistical parameters required for reproducing the measured directional spectra by means of Jonswap and cosine2s models are provided. The system was commissioned and a 9-month validation campaign was executed. It is shown that the estimated parameters show quite reasonable correlation with satellite-based hindcasts from NOAA for sea states with peak periods larger than 8sec.© 2013 ASME

A Shuttle Tanker Position Cooperative Control Applied to Oil Transfer Operations Based on the LQG/LTR Method

Abstract The present paper addresses the development of a cooperative control applied to Ships with Dynamic Position Systems (DP) under oil transfer operation. The objective of the cooperative control is to maintain constant the relative distance between the two shuttle tankers. The benefits of this control will be evaluated by numerical simulation. The obtained results will be compared to the non cooperative control.

Development of an On-Board Wave Estimation System Based on the Motions of a Moored FPSO: Commissioning and Preliminary Validation

This paper addresses the development, installation and initial tests of a system for wave spectra estimation from the measurements of the first order motions of a moored FPSO located in Campos Basin, Brazil. The estimation is based on Bayesian inference algorithms, previously validated by means of numerical and small-scale experimental analysis. A 6-dof inertial measurement unit (IMU) is used for monitoring the motions of the platform, and this information is sent to a remote data-base, also accessed by the wave-estimation system. The algorithm also requires the Response Amplitude Operators (RAOs), and they depend on the loading conditions of the FPSO.A previous analysis considering typical loading configurations of the tanks showed that the wave estimation is mainly dependent on the total displacement of the vessel, and not on the load distribution among the tanks. Hence, the RAOs for the full-range of drafts (or total displacements) were numerically generated, considering a uniform distribution of the load among the tanks. Since the draft of the platform was not directly measured, the loading levels of the tanks are obtained from the automation system of the platform, and the draft is then estimated. Finally, the heading is measured by a gyrocompass, and it is necessary for the definition of the global wave direction. The Bayesian estimation is executed at time-spans of 30min. A parametric optimization algorithm is then applied for the calculation of the wave spectrum parameters from the raw-spectrum obtained by the Bayesian estimation.A user-friendly interface was also developed, with on-line information about platform motions, estimated wave spectrum, peak statistics and data history. Since all information is accessed by network, the wave system can be installed either on-board or in the on-shore monitoring center.The system was commissioned and a partial 3-month validation campaign was executed. The spectrum results were compared to NOAA estimates. As expected, low-period wave components (smaller than 8s) could not be estimated with accuracy, since the FPSO presents small motion response for these components. Swell and high-period wave components estimates presented good qualitative and quantitative agreement with satellite prediction.Copyright

Longitudinal Control of Pirajuba Autonomous Underwater Vehicle, Using Techniques of Robust Control LQG/LTR

Abstract This work presents a study on a robust controller design for the motion on the vertical plane of an AUV at low depths. The LQG/LTR control strategy is proposed for rejecting disturbances originated from sea waves on the depth, attitude and axial velocity of an AUV. The vehicle is an adapted version of the Pirajuba AUV, developed at the Unmanned Vehicles Laboratory, in the University of Sao Paulo. The plant modeling was based on previous studies on manoeuvrability predictions for the AUV, and a simplified model for wave disturbances. Modeling errors for manoeuvring coefficients were considered in the controller design. The validation is based on simulation studies using a set of non-linear equations of motion.