Tomas Martinez-Marin

A Particle Filter Approach for InSAR Phase Filtering and Unwrapping

This paper presents a new phase-unwrapping (PU) algorithm for SAR interferometry that makes use of a particle filter (PF) to perform simultaneously noise filtering and PU. The formulation of this technique provides independence from noise statistics and is not constrained by the nonlinearity of the problem. In addition, an enhanced variant of this method combining a PF with artificial-intelligence search strategies and an omnidirectional local phase estimator, based on the mode of the power spectral density, is also presented. Results obtained with synthetic and real data show a significant improvement with respect to other conventional unwrapping algorithms in some situations.

Fast Reinforcement Learning for Vision-guided Mobile Robots

This paper presents a new reinforcement learning algorithm for accelerating acquisition of new skills by real mobile robots, without requiring simulation. It speeds up Q-learning by applying memory-based sweeping and enforcing the “adjoining property”, a technique that exploits the natural ordering of sensory state spaces in many robotic applications by only allowing transitions between neighbouring states. The algorithm is tested within an image-based visual servoing framework on a docking task, in which the robot has to position its gripper at a desired configuration relative to an object on a table. In experiments, we compare the performance of the new algorithm with a hand-designed linear controller and a scheme using the linear controller as a bias to further accelerate the learning. By analysis of the controllability and docking time, we show that the biased learner could improve on the performance of the linear controller, while requiring substantially lower training time than unbiased learning (less than 1 hour on the real robot).

Crop Phenology Estimation Using a Multitemporal Model and a Kalman Filtering Strategy

In this letter, a new approach for crop phenology estimation with remote sensing is presented. The proposed methodology is aimed to exploit tools from a dynamical system context. From a temporal sequence of images, a geometrical model is derived, which allows us to translate this temporal domain into the estimation problem. The evolution model in state space is obtained through dimensional reduction by a principal component analysis, defining the state variables, of the observations. Then, estimation is achieved by combining the generated model with actual samples in an optimal way using a Kalman filter. As a proof of concept, an example with results obtained with this approach over rice fields by exploiting stacks of TerraSAR-X dual polarization images is shown.

Estimation of Key Dates and Stages in Rice Crops Using Dual-Polarization SAR Time Series and a Particle Filtering Approach

Information of crop phenology is essential for evaluating crop productivity. In a previous work, we determined phenological stages with remote sensing data using a dynamic system framework and an extended Kalman filter (EKF) approach. In this paper, we demonstrate that the particle filter is a more reliable method to infer any phenological stage compared to the EKF. The improvements achieved with this approach are discussed. In addition, this methodology enables the estimation of key cultivation dates, thus providing a practical product for many applications. The dates of some important stages, as the sowing date and the day when the crop reaches the panicle initiation stage, have been chosen to show the potential of this technique.

Dynamical Approach for Real-Time Monitoring of Agricultural Crops

In this paper, a novel approach for exploiting multitemporal remote sensing data focused on real-time monitoring of agricultural crops is presented. The methodology is defined in a dynamical system context using state-space techniques, which enables the possibility of merging past temporal information with an update for each new acquisition. The dynamic system context allows us to exploit classical tools in this domain to perform the estimation of relevant variables. A general methodology is proposed, and a particular instance is defined in this study based on polarimetric radar data to track the phenological stages of a set of crops. A model generation from empirical data through principal component analysis is presented, and an extended Kalman filter is adapted to perform phenological stage estimation. Results employing quad-pol Radarsat-2 data over three different cereals are analyzed. The potential of this methodology to retrieve vegetation variables in real time is shown.

Using a Grid-Based Filter to Solve InSAR Phase Unwrapping

This letter presents a phase-unwrapping (PU) algorithm for synthetic aperture radar interferometry based on a grid-based filter. The proposed PU algorithm, which is based on state-space techniques, simultaneously performs noise filtering and PU. The formulation of this technique provides independence from noise statistics and is not constrained by the nonlinearity of the problem. Results obtained with synthetic data show a significant improvement with respect to other conventional PU algorithms in some situations.

Optimal motion planning by reinforcement learning in autonomous mobile vehicles

The aim of this work has been the implementation and testing in real conditions of a new algorithm based on the cell-mapping techniques and reinforcement learning methods to obtain the optimal motion planning of a vehicle considering kinematics, dynamics and obstacle constraints. The algorithm is an extension of the control adjoining cell mapping technique for learning the dynamics of the vehicle instead of using its analytical state equations. It uses a transformation of cell-to-cell mapping in order to reduce the time spent during the learning stage. Real experimental results are reported to show the satisfactory performance of the algorithm.

An Optimized Algorithm for InSAR Phase Unwrapping Based on Particle Filtering, Matrix Pencil, and Region-Growing Techniques

This letter presents a new phase unwrapping algorithm for synthetic aperture radar interferometry which combines a particle filter, a matrix-pencil (MP) local slope estimator, and an optimized region-growing technique. The advantages of the new method rely on the following contributions: The MP estimator provides better resolution to the local slope estimation, the particle filter enables simultaneous unwrapping and filtering without a priori statistics constraints, and the implemented region-growing technique adds diversity of unwrapping paths and ensures an optimum solution. Results introduced in this letter illustrate the main aspects of the new approach.

On-line optimal motion planning for nonholonomic mobile robots

In this paper we propose a novel approach for on-line motion planning of nonholonomic robots through reinforcement learning. The algorithm incorporates a mechanism, the adjoining property, to select the state transitions that will be learned by the robot controller. The method overcomes some limitations of reinforcement learning techniques when they are employed in applications with continuous non-linear systems, such as nonholonomic vehicles. Furthermore, a good approximation to the optimal behaviour is obtained by a look-up table without of using function interpolation. Finally, we present both simulation and experimental results to show the satisfactory performance of the method compared with the popular Q-learning algorithm

Navigation of Autonomous Vehicles in Unknown Environments using Reinforcement Learning

In this paper we propose a generic approach for navigation of nonholonomic vehicles in unknown environments. The vehicle model is also unknown, so the path planner uses reinforcement learning to acquire the optimal behaviour together with the model, which is estimated by a reduced set of transitions. After the training phase, the vehicle is able to explore the environment through a wall-following behaviour. In order to guide the navigation and to build a map of the environment the planner employs virtual walls. The learning time to acquire a good approximation of the wall-following behaviour was only a few minutes. Both simulation and experimental results are reported to show the satisfactory performance of the method.