Javier Gimenez

Optimization methodology to fruit grove mapping in precision agriculture

A method capable of efficiently mapping a semi-structured environment is presented.Grove mapping based on LiDAR and the GPS locations of the corner trees is given.An optimization tool that adjusts measurements acquired by a mobile robot is used.The technique was tested in an olive grove located in San Juan - Argentina.It is incorporated a novel filtering technique of unlikely data. The mapping of partially structured agricultural environments is a valuable resource for precision agriculture. In this paper, a technique for the mapping of a fruit grove by a mobile robot is proposed, which uses only front laser information of the environment and the exact position of the grove corners. This method is based on solving an optimization problem with nonlinear constraints, which reduces errors inherent to the measurement process, ensuring an efficient and precise map construction. The resulting algorithm was tested in a real orchard environment. For this, it is also developed a data filtering method capable to comply efficiently the observation-feature matching. The maximum average error obtained by the methodology in simulations was about 13cm, and in real experimentation was about 36cm.

Inference strategies for the smoothness parameter in the Potts model

The Potts model is a commonplace in Bayesian image analysis since its introduction as a convenient image prior. It is able to describe the distribution of classes, yielding a regularization term in the cost function to be minimized in many classification problems. The simplest isotropic version depends on a scalar smoothness parameter; its value controls the relative influence of the regularization with respect to the data. This work analyzes the performance of two pseudolike-lihood estimation procedures of the smoothness parameter of the Potts model: the classical one, which employs the map of classes, and a new estimator based on the posterior distribution, which also incorporates the evidence provided by the observed data. Our simulation study shows that the combination of prior information and observation data gives accurate β estimations when true data is provided. We also discuss its influence in the classification results when comparing contextual ICM (Iterated Conditional Modes) classification experiments with multispectral optical imagery, estimating the scalar parameter β with our estimator and the classical one. Our experiment shows promising results, since ICM with our estimator is able to distinguish image features that the classical ICM does not.

Human-robot interaction in precision agriculture: Sharing the workspace with service units

Currently, Chile and Argentina experience serious challenges that affect their agricultural productivity. For example, in Chile, the loss of farmable field due to recent earthquakes, and volcanic eruptions as well as the loss of water reserves due to climate changes are affecting the agriculture. Additionally, both countries are facing a same problem: the loss of human labor force. Field workers are migrating from the farm to other fields in the industry which offer them more stable and more profitable jobs (like the mining industry in Chile, or car assembling lines in Argentina). In this adverse scenario, it becomes necessary to introduce and develop agricultural automation and sensing technologies for both primary (harvesting, seeding, fertilizing, spraying) and secondary tasks (grove supervision, weed detection, hauling, mowing). However, fully robotized farms are not yet a possibility since the transition from human labor force dependent farming to autonomous farming needs to be smooth and requires legal regulation not yet in discussion. In this paper, we summarize the state of the art in human-robot interaction in farmable fields, with emphasis in the current constrains associated with flexible automatization of farms in Argentina and Chile. In particular, we introduce the guidelines for designing a humanrobot interaction strategy for harvesting tasks, that could be used for other agricultural tasks.

When Data Do Not Bring Information: A Case Study in Markov Random Fields Estimation

The Potts model is frequently used to describe the behavior of image classes, since it allows to incorporate contextual information linking neighboring pixels in a simple way. Its isotropic version has only one real parameter β, known as smoothness parameter or inverse temperature, which regulates the classes map homogeneity. The classes are unavailable and estimating them is central in important image processing procedures as, for instance, image classification. Methods for estimating the classes which stem from a Bayesian approach under the Potts model require to adequately specify a value for β. The estimation of such parameter can be efficiently made solving the pseudo maximum-likelihood (PML) equations in two different schemes, using the prior or the posterior model. Having only radiometric data available, the first scheme needs the computation of an initial segmentation, whereas the second uses both the segmentation and the radiometric data to make the estimation. In this paper, we compare these two PML estimators by computing the mean-square error (MSE), bias, and sensitivity to deviations from the hypothesis of the model. We conclude that the use of extra data does not improve the accuracy of the PML; moreover, under gross deviations from the model, this extra information introduces unpredictable distortions and bias.

A New Approach to Image Segmentation with Two-Dimensional Hidden Markov Models

Image segmentation is one of the fundamental problems in computer vision. In this work, we present a new segmentation algorithm that is based on the theory of two-dimensional hidden Markov models (2D-HMM). Unlike most 2D-HMM approaches we do not apply the Viterbi Algorithm, instead we present a computationally efficient algorithm that propagates the state probabilities through the image. This approach can easily be extended to higher dimensions. We compare the proposed method with a 2D-HMM standard algorithm and Iterated Conditional Modes using real world images like a radiography or a satellite image as well as synthetic images. The experimental results show that our approach is highly capable of condensing image segments. This gives our algorithm a significant advantage over the standard algorithm when dealing with noisy images with few classes.

A New Approach to Segmentation of Multispectral Remote Sensing Images Based on MRF

Segmentation of multispectral remote sensing images is a key competence for a great variety of applications. Many of the applied segmentation algorithms are generative models based on Markov random fields. These approaches are generally limited to multivariate probability densities such as the normal distribution. In addition, it is usually impossible to adjust the contextual parameters separately for each frequency band. In this letter, we present a new segmentation algorithm that avoids the aforementioned problems and allows the use of any univariate density function as emission probability in each band. The approach consists of three steps: first, calculate feature vectors for every frequency band; second, estimate contextual parameters for every band and apply local smoothing; and third, merge the feature vectors of the frequency bands to obtain final segmentation. This procedure can be iterated; however, experiments show that after the first iteration, most of the pixels are already in their final state. We call our approach successive band merging (SBM). To evaluate the performance of SBM, we segment a Landsat 8 and an AVIRIS image. In both cases, the k̂ coefficients show that SBM outperforms the benchmark algorithms.

Multi-objective control for cooperative payload transport with rotorcraft UAVs

A novel kinematic formation controller based on null-space theory is proposed to transport a cable-suspended payload with two rotorcraft UAVs considering collision avoidance, wind perturbations, and properly distribution of the load weight. An accurate 6-DoF nonlinear dynamic model of a helicopter and models for flexible cables and payload are included to test the proposal in a realistic scenario. System stability is demonstrated using Lyapunov theory and several simulation results show the good performance of the approach.

Bounded memory probabilistic mapping of out-of-structure objects in fruit crops environments

Abstract Spatial awareness and memory are key factors for a robot to evolve in semi-structured and dynamic environments as those found in agriculture, and particularly in fruit crops where the trees are regularly distributed. This paper proposes a probabilistic method for mapping out-of-structure objects (weeds, workers, machines, fallen branches, etc.) using a Kernel density estimator. The methodology has theoretical and practical advantages over the well-known occupancy grid map estimator such as optimization of storage resources, online update, high resolution, and straightforward adaptability to dynamic environments. An example application would be a control scheme through which a robot is able to perform cautious navigation in areas with high probability of finding obstacles. Simulations and experiments show that large extensions can be online mapped with few data and high spatial resolution.

Port-Hamiltonian modelling of a car-like robot

This paper proposes a dynamical modelling of a car-like robot based on the port-Hamiltonian approach. It consists in projecting the dynamics of a free body into the possible velocity space determined by the non-holonomic constraints of the vehicle. For this approach, it is considered a simplified bicycle-like representation of the Ackermann mechanism with rear traction and steering control on the front wheel. Simulations are given to illustrate the effectiveness of the approach.

Port-hamiltonian modelling of a differential drive mobile robot with reference velocities as inputs

In this paper it is proposed a dynamic model of a differential drive mobile robot based on the port-Hamiltonian approach. The model inputs are reference velocities, as in most commercial robots. The model arises from a positioning controller at desired velocities, which is asymptotically stable.