**3. Automation of conventional vehicles**

The tractor has been the central vehicle for executing most of the work required in a crop field. Equipped with the proper accessories, this machine can till, plant, fertilize, spray, haul, mow, and even harvest. Their adaptability to dissimilar tasks

#### **Figure 2.**

*An example of agricultural tractor automation-distribution of sensorial and actuation systems for transforming an agricultural tractor into a UGV (Gonzalez-de-Santos et al., 2017).*

makes tractors a prime target for automation, which would enable productivity increases, improve safety, and reduce operational costs. **Figure 2** shows an example of the technologies and equipment for automating agricultural tractors.

Numerous worldwide approaches to automating diverse types of tractors have been researched and developed since 1995 when the first GNSS was made available to the international civilian community of users, which opened the door for GPSguided agricultural vehicles (auto-steering) and controlled-traffic farming.

The first evaluations of GPS systems for vehicle guidance in agriculture were also published in 1995 [14] demonstrating its potential and encouraging many research groups around the world to automate diverse types of tractors. The earliest attempts were made at Stanford University in 1996, where an automatic control system for an agricultural tractor was developed and tested on a large farm [15]. The system used a location system with four GPS antennas. Around the same time, researchers at the University of Illinois, USA, developed a guidance system for an autonomous tractor based on sensor fusion that included machine vision, real-time kinematics GPS (RTK-GPS), and a geometric direction sensor (GDS). The fusion integration methodology was based on an extended Kalman filter (EKF) and a twodimensional probability-density-function statistical method. This system achieved a lateral average error of approximately 0.084 m at approximately 2.3 m s<sup>−</sup><sup>1</sup> [16].

A few years later, researchers at Carnegie Mellon University, USA, developed some projects that made significant contributions. The Demeter project was conceived as a next-generation self-propelled hay harvester for agricultural operations, and it became the most representative example of such activity [17]. The positional data was fused from a differential GPS, a wheel encoder (dead reckoning), and gyroscopic system sensors. The project resulted in a system that allowed an expert harvesting operator to harvest a field once, thus programming the field. Subsequently, an operator with lesser skill could "playback" the programmed field at a later date. The semi-autonomous agricultural spraying project, developed by the same research group, was devoted to making pesticide spraying significantly cheaper, safer, and more environmentally friendly [18]. This system enabled a remote operator to oversee the nighttime operation of up to four spraying vehicles. Another example is research conducted at the University of Florida, USA, [19], in

**77**

**Table 1.**

*Unmanned Ground Vehicles for Smart Farms DOI: http://dx.doi.org/10.5772/intechopen.90683*

error of approximately 0.025 m.

Carnegie Mellon University (USA)—Demeter project [17]

Carnegie Mellon University (USA)—Autonomous Agricultural

Spraying project [18]

LASMEA-CEMAGREF (France) [20]

University of Aarhus and the University of Copenhagen (Denmark) [21]

Carnegie Mellon University

*UGVs based on commercial vehicles.*

(USA) [24]

which two individual autonomous guidance systems for use in a citrus grove were developed and tested along curved paths at a speed of approximately 3.1 m s<sup>−</sup><sup>1</sup>

system, based on machine vision, achieved an average guidance error of approximately 0.028 m. The other system, based on LIDAR guidance, achieved an average

Similar activities started in Europe in the 2000s. One example is the work performed at LASMEA-CEMAGREF, France, in 2001, which evaluated the possibilities of achieving recording-path tracking using a carrier phase differential GPS (CP-DGPS), as the only sensor. The vehicle heading was derived according to a Kalman state reconstructor and a nonlinear velocity independent control law was

A relevant example of integrating UGVs with automated tools is the work conducted at the University of Aarhus and the University of Copenhagen, Denmark [21]. The system comprised an autonomous ground vehicle and a side shifting arrangement affixed to a weeding implement. Both the vehicle and the implement were equipped with RTK-GPS; thus, the two subsystems provided their own positions, allowing the vehicle to follow predefined GPS paths and enabling the implement to act on each individual plant, whose positions were automatically obtained during seeding.

Lately, some similar automations of agricultural tractors have been conducted using more modern equipment [22, 23], and some tractor manufacturers have already presented noncommercial autonomous tractors [11, 12]. This tendency to automate existing tractors has been applied to other types of lightweight vehicles for specific tasks in orchards such as tree pruning and training, blossom and fruit thinning, fruit harvesting, mowing, spraying, and sensing [24]. **Table 1** summarizes

Stanford University (USA) [15] 1996 Automatic large-farm tractor using 4 GPS antennas University of Illinois (USA) [16] 1998 A guidance system using a sensor based on machine

University of Florida (USA) [19] 2006 An autonomous guidance system for citrus groves

RHEA consortium (EU) [22] 2014 A fleet (3 units) of tractors that cooperated and

University of Leuven (Belgium) [23] 2015 Tractor guidance using model predictive control for

vision, an RTK-GPS, and a GDS

operations

vehicle guiding using a CP-DGPS as the only sensor

based on machine vision and LADAR

connected to an unmanned tractor

collaborated in physical/chemical weed control and pesticide applications for trees

yaw dynamics

1999 A self-propelled hay harvester for agricultural

2002 A ground-based vehicles for pesticide spraying

2001 This study investigated the possibility of achieving

2008 An automatic intra-row weed control system

2015 Self-driving orchard vehicles for orchard tasks

designed that relied on chained systems properties [20].

the UGVs based on commercial vehicles for agricultural tasks.

**Institution Year Description**

. One

#### *Unmanned Ground Vehicles for Smart Farms DOI: http://dx.doi.org/10.5772/intechopen.90683*

*Agronomy - Climate Change and Food Security*

makes tractors a prime target for automation, which would enable productivity increases, improve safety, and reduce operational costs. **Figure 2** shows an example

*An example of agricultural tractor automation-distribution of sensorial and actuation systems for* 

*transforming an agricultural tractor into a UGV (Gonzalez-de-Santos et al., 2017).*

Numerous worldwide approaches to automating diverse types of tractors have been researched and developed since 1995 when the first GNSS was made available to the international civilian community of users, which opened the door for GPSguided agricultural vehicles (auto-steering) and controlled-traffic farming.

The first evaluations of GPS systems for vehicle guidance in agriculture were also published in 1995 [14] demonstrating its potential and encouraging many research groups around the world to automate diverse types of tractors. The earliest attempts were made at Stanford University in 1996, where an automatic control system for an agricultural tractor was developed and tested on a large farm [15]. The system used a location system with four GPS antennas. Around the same time, researchers at the University of Illinois, USA, developed a guidance system for an autonomous tractor based on sensor fusion that included machine vision, real-time kinematics GPS (RTK-GPS), and a geometric direction sensor (GDS). The fusion integration methodology was based on an extended Kalman filter (EKF) and a twodimensional probability-density-function statistical method. This system achieved

a lateral average error of approximately 0.084 m at approximately 2.3 m s<sup>−</sup><sup>1</sup>

A few years later, researchers at Carnegie Mellon University, USA, developed some projects that made significant contributions. The Demeter project was conceived as a next-generation self-propelled hay harvester for agricultural operations, and it became the most representative example of such activity [17]. The positional data was fused from a differential GPS, a wheel encoder (dead reckoning), and gyroscopic system sensors. The project resulted in a system that allowed an expert harvesting operator to harvest a field once, thus programming the field. Subsequently, an operator with lesser skill could "playback" the programmed field at a later date. The semi-autonomous agricultural spraying project, developed by the same research group, was devoted to making pesticide spraying significantly cheaper, safer, and more environmentally friendly [18]. This system enabled a remote operator to oversee the nighttime operation of up to four spraying vehicles. Another example is research conducted at the University of Florida, USA, [19], in

[16].

of the technologies and equipment for automating agricultural tractors.

**76**

**Figure 2.**

which two individual autonomous guidance systems for use in a citrus grove were developed and tested along curved paths at a speed of approximately 3.1 m s<sup>−</sup><sup>1</sup> . One system, based on machine vision, achieved an average guidance error of approximately 0.028 m. The other system, based on LIDAR guidance, achieved an average error of approximately 0.025 m.

Similar activities started in Europe in the 2000s. One example is the work performed at LASMEA-CEMAGREF, France, in 2001, which evaluated the possibilities of achieving recording-path tracking using a carrier phase differential GPS (CP-DGPS), as the only sensor. The vehicle heading was derived according to a Kalman state reconstructor and a nonlinear velocity independent control law was designed that relied on chained systems properties [20].

A relevant example of integrating UGVs with automated tools is the work conducted at the University of Aarhus and the University of Copenhagen, Denmark [21]. The system comprised an autonomous ground vehicle and a side shifting arrangement affixed to a weeding implement. Both the vehicle and the implement were equipped with RTK-GPS; thus, the two subsystems provided their own positions, allowing the vehicle to follow predefined GPS paths and enabling the implement to act on each individual plant, whose positions were automatically obtained during seeding.

Lately, some similar automations of agricultural tractors have been conducted using more modern equipment [22, 23], and some tractor manufacturers have already presented noncommercial autonomous tractors [11, 12]. This tendency to automate existing tractors has been applied to other types of lightweight vehicles for specific tasks in orchards such as tree pruning and training, blossom and fruit thinning, fruit harvesting, mowing, spraying, and sensing [24]. **Table 1** summarizes the UGVs based on commercial vehicles for agricultural tasks.


#### **Table 1.**

*UGVs based on commercial vehicles.*

Nevertheless, UGVs suitable for agriculture remain far from commercialization, although many intermediate results have been incorporated into agricultural equipment—from harvesting to precise herbicide application. Essentially, these systems are installed on tractors owned by farmers and generally consist of a computer (the controller), a device for steering control, a localization system (mostly based on RTK-GPS), and a safety system (mostly based on LIDAR). Many of these systems are compatible only with advanced tractors that feature ISOBUS control technology [25], through which controllers connected to the ISOBUS can access other subsystems of the tractor (throttle, brakes, auxiliary valves, power takeoff, linkage, lights, etc.). Examples of these commercial systems are AutoDrive [26] and X-PERT [27].

An important shortcoming of these solutions is their lack of intelligence in solving problems, especially when obstacles are detected because they are not equipped with technology suitable for characterizing and identifying the obstacle type. This information is essential when defining any behavior other than simply stopping and waiting for the situation to be resolved. Another limitation of this approach is that the conventional configuration of a standard tractor driven by an operator is designed to maximize the productivity per hour; thus, the general architecture of the system (tractor plus equipment) is only roughly optimized.
