Robotic Pesticide-free Control of Pests in Agriculture
Arnaud Lelevé, Pedro Da Silva, Vincent Dufour, François Feugier, Robin Gaetani, Abdelaziz Heddi, BrunoMasenelli, Virginie Lacotte, Richard Moreau, Minh Tu Pham, KantyRabenorosoa of BF2I, Ampère, GST, INL and FEMTO-ST.
Due to the recent informatic and robotic advances and like many other fields (industry, medicine, education), the modern agriculture is also in undergoing technological revolution. The enlargement of farms, the lack of workers (even more critical in time of health crisis), the ecological politics imposing a reduction of phytosanitary products, produce an unprecedented pressure on the agricultural world, pushing it to rethink. Thus, the agricultural mechanization is giving away the place to robotization. Among the current needs in agricultural robotics, the detection of illness or pest and their targeted destruction have been not much studied . Yet an early detection can limit their impact on the production, and a targeted destruction has no impact on the environment. The GreenShield project aims to reduce the use of pesticides by developing a robotized module onboard of a vehicle (mobile robot or tractor) to fight crops pests. The robot will patrol in the crops to scan the young plants of invertebrate research  and of illness symptoms . It will thus collect reliable data in terms of nature and contamination rate, who will serve to optimize the following campaigns. As soon it detects the pests, it will neutralize with the help of a guiding laser beam. A solution already studied for the fight against weeds  and industrialize by the startup Green Shield Technology (GST).
This project, started in 2017, implements the skills of 4 research laboratories: animal biology (BF2I), laser photonics (INL), micro-systems (Femto-ST) and robotic (Ampère), at the service of a start-up (Green Shield Technology). To date, the detection and neutralization functions have been validated individually. As regards the detection, reflectivity spectra were measured on three species of aphids and one cryptogamic fungus: the mildew of the vineyard. The results show the possibility of detecting different aphids’ species on their spectral signature base between 400 and 1000 nm and the vineyard infection by mildew in a laboratory before the appearance of symptoms visible to the necked eyes. Currents experimentation have demonstrated the feasibility of detecting the position of aphids in space, thanks to a stereo vision camera only in the visible spectrum. An Artificial Intelligence algorithm has been trained for this purpose. An optimization is still necessary to get rid of the variability of the natural lighting. Concerning the neutralization, several types of laser sources have been tested. We converged on blue and green solid-state lasers of 10 W power, weighing only a few kilograms. The lethality tests of these lasers allowed us to determinate a necessary fluence of 11 J/cm² to obtain 90% mortality at D+1 for aphids (all ages combined). The laser orientation is realized by a commercial micromirror. Tests aiming in laboratories showed a sub-millimeter open-loop precision when the aphids were in the center of the image. A work of visual enslavement is essential to enlarge this zone. Nowadays, a first unidirectional prototype of a mobile robot, rolling above of the bean plants manually colonized by aphids, has allowed to validate the detection during rolling. A second prototype is under realization: more manageable, with less vibration and integrating the neutralization laser, it will be the support of the next experimentation integrating the whole order chain.
Multi-trajectory approach for a generic coordination paradigm of a wheeled mobile manipulator
Today, many challenges are facing the agricultural sector, not only in production but also in environmental protection and society’s concern. Robotics introduction in agricultural environments can provide an answer to these different issues; it can help to reduce environmental damages caused by the very intensive production and facilitate the farmer’s work while keeping their productivity strong. A lot of research project has been investigating for many different tasks such as weeding, crop monitoring, treatment, or harvesting concerning all agricultural area as vineyard, greenhouse, or field. However, this agricultural robotization faced a major issue: task flexibility. Indeed, all commercialized robots are designed for a single task. However, if such systems are multiplied by the number of tasks to be performed, the costs of them are too important for a farmer. To address this problem, it is essential to develop autonomous systems that can perform a variety of tasks. A possible way to increase robot flexibility is to offer them multiple action modalities with a system that possesses mobilities and manipulation capacities. Such systems are generally called a mobile manipulator, but agricultural applications are mainly interested in the wheeled mobile manipulator. Since 2000, many research works have been done on wheeled mobile manipulators to address critical challenges of this robot: coordination control and redundancy resolution. Two important ways to resolve this issue has appeared through literature: internally coupled approaches and externally coupled approaches.
The first one proposes to model and control system with single modeling and control law. In this approach, it is simple to take into account of redundancy degree of the system to resolve them. However, modeling and controlling is complicated because the control space becomes larger and can be subject to non-holonomic constraint. A second way is to control each subsystem by simple models and control laws already existing in the literature and coordinated all control points of each subsystem in a high-level coordination stage. With this configuration, modeling and control are very simple, but redundancy resolution is hard because there is no global modeling which takes account of all redundancy degree. The trend in current research today is towards internally coupled approaches. However, the latter suffers from an important flaw: the genericity of their approach. Indeed, for each different structure of robots, models and control laws must have to be redesigned. These approaches are not adjusted to manage various types of robots or adaptive structures.
From this observation, it follows that the presented work proposes an externally coupled approach to create a generic solution facing the diversity of robot structures. Also, opting for different ways of control for each subsystem permit differentiation in their types. Indeed, in a natural environment with varying and uncertain dynamics, a kinematic control for the moving base is relevant when a dynamic control of the manipulator is more suitable for gripping tasks. This external coupling approach then proposes a generic paradigm for the coordination based on existing modeling and control laws. To do this, the coordination stage calculates the variable deviations along each reference trajectory associated with a subsystem before sending them to the control stage in charge of regulating each subsystem on these different set points. The calculation of these set points is done by a gradient- based optimization of a coordination function into the coordination stage. This function is formed with several weighted criteria. These criteria allow, according to their nature and expression, to give a coherent global behavior to the system. The use of gradient-based optimization allows changing criteria according to the final application while keeping the coordination stage independent of the robot used. The proposed approach showed its effectiveness on an experimental platform for effector tracking the trajectory in anticipation of performing a treatment along a vegetation row.