Agricultural robots, a laboratory in the field

On 25 and 26 May, in Cornaredo (MI), the Cascina Baciocca farm hosted the ACRE event, dedicated to robots and highly automated means for agriculture. The event, organised by Davide Facchinetti of the Department of Agricultural and Food Sciences of the University of Milan in collaboration with Matteo Matteucci of the Milan Polytechnic, offered a preview of what agriculture will be like in the near future

The edition in Cornaredo is the second for the ACRE competition (acronym for Agri-food Competition for Robot Evaluation), while the first was held in Montoldre, France in June 2022. At the event, around a dozen robots from all over the world tackled specific tasks in the field of agriculture. The event was attended by around 500 people, represented - for the most part - by students attending the area's agricultural technical institutes, as well as a number of leading representatives of the best-known agricultural machinery manufacturers, who were also particularly impressed by the spirit in which the initiative was carried out. As Matteo Matteucci explains, although it was a competition, with official judges and precise rules, the real objective for the teams was not to win, but rather to test their prototypes in a standardised real field situation and, above all, to meet and debate with other researchers working in the same field, exchanging information and advice, in order to collaborate on improving the technical solutions and Artificial Intelligence algorithms used on the various robots.

This event shares the same aim as that of a very related project called 'Agrifood TEF', which will start soon and is also organised by the University of Milan in collaboration with the Milan Polytechnic and with the involvement of Federunacoma. This aim, in the opinion of Davide Facchinetti, is to bring Italian agricultural machinery manufacturers as close as possible to the world of research carried out in the academic sphere regarding the use of robotics and artificial intelligence in agriculture. In the United States, France, as well as Denmark, Holland and Germany, hundreds of millions of euros have already been invested in research in this specific sector over the last few years. This has allowed them to produce many prototypes, as well as to market several robotic solutions. It is therefore fundamental for Italy to find solutions that will allow it to bridge the existing gap that separates us, giving support to the national agricultural machinery industry to develop robotic solutions. This is a sector that will certainly expand very rapidly in the near future, thanks to the great progress made by Artificial Intelligence in recent years. It will allow robots to be used even in environments such as agriculture, characterised by an extremely greater variability of situations and scenarios than in any industrial environment, where robots are now the masters.

According to Matteo Matteucci, the decrease in the prices of optical sensors such as time-of-flight and multispectral cameras, as well as radar and LIDAR, will make robots increasingly affordable in the short term, although there is no doubt that their deployment will start in the most profitable crops.

The ACRE competition was structured in such a way as to allow participants to test their abilities on different types of crops (specifically on this occasion maize, bean and chard) by performing specific tasks. For example, some robots, using vision and artificial intelligence techniques, determined which plants in the 'investigated' area were the cultivated ones and which were the weeds. Others navigated autonomously between the rows of the crop, deliberately sown following non-straight paths, without damaging it, and still other robots could perform both tasks simultaneously, as well as removing weeds.

In order to broaden the landscape of robotic solutions presented also outside the specific field of weed control, this year's ACRE competition was also coupled with a dynamic field exhibition of robots dedicated to other fields, so that this event saw both robots developed by various university start-ups and solutions proposed by companies already offering commercial products.

 

Agrovai

The Salerno-based company took part in the event with a robot in an advanced engineering stage, designed to carry out weed eradication in greenhouses intended for growing baby leaves (range IV salads). It uses vision and artificial intelligence to recognise weeds from the current crop and then removes the weeds with special tweezers and stores them in a special container. At the ACRE event, however, the robot was not equipped with this actuator, as operating on compacted soil and not on lettuce was unsuitable for weed control, but it demonstrated excellent navigation and weed recognition capabilities by discriminating the weeds from the crop being grown (maize and beans).

FederUnacoma member companies included Arvatec, which in addition to being the technical sponsor that provided the GPS RTK correction signal to all participants, took part with no less than three robots, and Barbieri, which participated with two robots.

 

Arvatec

The company participated in ACRE with both the Danish robot FarmDroid and Moondino, the robot designed and conceived by Arvatec itself. Both are guided using an RTK GPS and are equipped with photovoltaic panels and batteries to enable them to move using solar energy alone. While the former operates on weeded crops and also carries out mechanical weeding along the row using special hoes, the latter (also designed for operation in rice fields) uses several passes of its toothed wheels between the rows to contain weed growth. SlopeHelper, on the other hand, is a Slovenian robot marketed by Arvatec, similar to a tracked tool carrier dedicated to operating in vineyards and orchards using equipment fitted with specific attachments for this robot.

 

Barbieri

The Italian company took part in the ACRE event as an exhibitor, presenting to the public two mowing robots capable of operating well even on extreme slopes and in complete autonomy thanks to a centimetric satellite guidance system. The two solutions differed in their cutting widths and in the fact that while one was driven by a classic endothermic motor, the other employed a brushless electric motor which used a 10 kWh lithium-ion battery made by Barbieri itself.

 

British Columbia University

A group of undergraduate students at this prestigious university has developed a prototype robot with a quadrangular structure equipped with four drive wheels, two of which are steerable. The robot also uses an RGBD camera and artificial intelligence that allows it to navigate the cultivated field without damaging the crop in progress and to distinguish weeds in order to activate nozzles controlled by a very common commercial Raspberry Pi processor only in their presence.

 

Milan Polytechnic

The Polytechnic presented two robots named 'SCOUT' and 'Rockerbot'. The first is a modular platform on which various types of sensors can be mounted. On this occasion it was equipped with an RgbD camera for acquiring images with which to train Artificial Intelligence. This is a stereoscopic camera that can also see in infrared. Equipped with four non-steered driving wheels, the vehicle performs turns a bit like a crawler, i.e. independently managing the speed of the individual wheels. In addition to the RTK GPS, for navigation the robot also uses a 64-plane LIDAR, which allows it to operate even in conditions where there is no satellite signal.

Rockerbot is similar to its predecessor in terms of steering modes, but employs a special structure that allows it to better absorb unevenness in the terrain and improve traction in rough terrain. In terms of sensors, Rockerbot uses a 32-plane LiDAR coupled with a 1-plane laser scanner, while the optical signal is acquired by an Intel RealSense camera and an OAK camera with a neural inference chip to reduce the computational load to be carried out downstream. The robot was also designed to perform site-specific weed spraying, but at the moment the spraying is only simulated.

 

University of Milan

Also attending the event, but not involved in weed control competitions, a tracked rover based on MATTRO and prototyped by the University of Milan in collaboration with Infosolutions. In this case, it is a robot designed for greenhouse operations, equipped with Lidar and special cameras that enable it to recognise different types of plants, continuously monitor their development, and in the future also recognise certain diseases. The robot detects and monitors the daily growth of each individual plant in the greenhouse and sends the data to a remote database.

 

University of Milano-Bicocca

Developed by a team led by Domenico Sorrenti, the robot built by the University of Milano-Bicocca is based on a three-wheeled robotic platform that employs two driving wheels positioned on the same axle and a pivoting wheel. Entirely designed and built in the University's IRALab laboratory, it employs a modular system of optical sensors, and was also equipped with an RTK GPS for navigation in this competition.