Browsing by Author "Auat Cheein, Fernando"

Now showing 1 - 7 of 7
  • No Thumbnail Available
    Item
    A two-stage deep learning strategy for weed identification in grassfields
    (2024) Calderara Cea, Felipe Antonio; Torres Torriti, Miguel Attilio; Auat Cheein, Fernando; Delpiano, José
    Machine vision strategies for weed identification, whether in industrial crops or grassfields, are fundamental to the development of automated removal systems necessary to increase agricultural yield and field maintenance efficiency. Identifying plant species considered invasive on grassfields is particularly challenging due to reduced color and morphological contrast, as well as phenotypic variability. This work presents a two-stage weed identification strategy using visible spectrum images. The first stage employs a convolutional siamese neural network to identify candidate regions that may contain weeds of irregular or regular morphology. The second stage employs a convolutional neural network to confirm the presence of irregular morphology weeds. The results of each stage are combined to produce an output containing a per-pixel probability of irregular weed and bounding boxes for the morphologically regular weed. The two-stage strategy has an accuracy score of 97.16% and a balanced accuracy score of 89.94% and macro F1 score of 81.14%. In addition to the good performance scores obtained with the proposed approach, it is to be noted that the convolutional Siamese network allows achieving a good performance with a relatively small dataset compared to other strategies that employ data-intensive training phases for optimizing the convolutional neural networks. The results were obtained with a dataset of weeds that has been made publicly available, as well as the neural network models and associated computer code. The dataset contains samples Trifolium repens and Lectuca virosa on grass obtained with two different cameras under varying illumination conditions and different geographic locations. The lightweight nature of the proposed strategy provides a solution amenable to implementation using currently existing embedded computer technology for real-time weed detection.
  • No Thumbnail Available
    Item
    Adaptive Nonlinear MPC for Efficient Trajectory Tracking Applied to Autonomous Mining Skid-Steer Mobile Robots
    (IEEE, 2020) Prado, Alvaro Javier; Chávez, Danilo; Camacho, Oscar; Torres Torriti, Miguel Attilio; Auat Cheein, Fernando
    The heterogeneous nature of the navigation surface suggests adaptation capabilities in vehicle motion control to overcome the effects of the wheel-terrain interaction. In such scenario, this paper presents an integral adaptive control framework built upon a Nonlinear Moving Horizon Estimator and a Nonlinear Model Predictive Control scheme, under which the objective is to on-line estimate states and model parameters of a robot motion model while autonomously navigating in off-road terrain conditions. With an adjustable model, the controller is made adaptive against terrain changes while tracking prescribed trajectories. The system is composed by two coupled subsystems to represent the vehicle motion and tire slip dynamics. The combined control-estimation strategy works under the Real-Time Iteration scheme to attain reliable computational activity for high-speed tire dynamics (e.g., slip). Trials in a simulation and real test environment with a compact mini-loader Cat® 262C, as those found in the mining industry, showed that the approach is able to efficiently estimate states and model parameters without exceeding constraints. The analysis of computational efficiency in various hardware configurations is also provided, exhibiting that the rapid optimization involved in the proposed controller is possible for high-speed dynamics.
  • Thumbnail Image
    Item
    Cluster Analysis for Agriculture
    (Springer, 2023) Arévalo Ramírez, Tito; Auat Cheein, Fernando
  • Thumbnail Image
    Item
    General Dynamic Model for Skid-Steer Mobile Manipulators with Wheel-Ground Interactions
    (2017) Aguilera Marinovic, Sergio Francisco; Torres Torriti, Miguel Attilio; Auat Cheein, Fernando
  • Thumbnail Image
    Item
    Overcoming the Loss of Performance in Unmanned Ground Vehicles Due to the Terrain Variability
    (2018) Prado, Javier; Yandun, Francisco; Torres Torriti, Miguel Attilio; Auat Cheein, Fernando
  • No Thumbnail Available
    Item
    Passive Landmark Geometry Optimization and Evaluation for Reliable Autonomous Navigation in Mining Tunnels Using 2D Lidars
    (2022) Torres-Torriti, Miguel; Nazate-Burgos, Paola; Paredes-Lizama, Fabian; Guevara, Javier; Auat Cheein, Fernando
    Autonomous navigation in mining tunnels is challenging due to the lack of satellite positioning signals and visible natural landmarks that could be exploited by ranging systems. Solutions requiring stable power feeds for locating beacons and transmitters are not accepted because of accidental damage risks and safety requirements. Hence, this work presents an autonomous navigation approach based on artificial passive landmarks, whose geometry has been optimized in order to ensure drift-free localization of mobile units typically equipped with lidar scanners. The main contribution of the approach lies in the design and optimization of the landmarks that, combined with scan matching techniques, provide a reliable pose estimation in modern smoothly bored mining tunnels. A genetic algorithm is employed to optimize the landmarks' geometry and positioning, thus preventing that the localization problem becomes ill-posed. The proposed approach is validated both in simulation and throughout a series of experiments with an industrial skid-steer CAT 262C robotic excavator, showing the feasibility of the approach with inexpensive passive and low-maintenance landmarks. The results show that the optimized triangular and symmetrical landmarks improve the positioning accuracy by 87.5% per 100 m traveled compared to the accuracy without landmarks. The role of optimized artificial landmarks in the context of modern smoothly bored mining tunnels should not be understated. The results confirm that without the optimized landmarks, the localization error accumulates due to odometry drift and that, contrary to the general intuition or belief, natural tunnel features alone are not sufficient for unambiguous localization. Therefore, the proposed approach ensures grid-based SLAM techniques can be implemented to successfully navigate in smoothly bored mining tunnels.
  • No Thumbnail Available
    Item
    Tube-based nonlinear model predictive control for autonomous skid-steer mobile robots with tire-terrain interactions
    (2020) Javier Prado, Alvaro; Torres-Torriti, Miguel; Yuz, Juan; Auat Cheein, Fernando
    This work addresses the problem of robust tracking control for skid-steer mobile platforms, using tube-based Nonlinear Model Predictive Control. The strategy seeks to mitigate the impact of disturbances propagated to autonomous vehicles originated by traction losses. To this end, a dynamical model composed by two coupled sub-systems stands for lateral and longitudinal vehicle dynamics and fire behavior. The controller is aimed at tracking prescribed stable operation points of the slip and side-slip beyond the robot pose and speeds. To reach robust tracking performance on the global system, a centralized control scheme operates under a predictive control framework composed by three control actions. The first one compensates for disturbances using the reference trajectory-feedforward control. The second control action corrects the errors generated by the modeling mismatch. The third one is devoted to ensure robustness on the closed-loop system whilst compensating for deviations of the state trajectories from the nominal ones (i.e. disturbance-free). The strategy ensures robust feasibility even when tightening constraints are met. Such constraints are calculated on-line based on robust positively invariant sets characterized by polytopic sets (tubes), including a terminal region to guarantee robustness. The benefits of the controller regarding tracking performance, constraint satisfaction and computational practicability were tested through simulations with a Cat (R) 262C skid-steer model. Then, in field tests, the controller evidenced high tracking accuracy against terrain disturbances when benchmarking performance with respect to inherent robust predictive controllers.