A Low-Cost Perception Improvement of an Electromechanical Gripper for Non-Destructive Fruit Harvesting ^†

Abstract

Modern intelligent robotic systems offer farmers a promising solution to labor shortages caused by socio-economic instability and/or pandemics. Efficient harvesting of delicate fruits is one of the main needs in this area. In this context, this work presents a simple and low-cost improvement of the ability of a servo-electric gripper to adjust its force when picking delicate fruits without damaging them. Specifically, this module utilizes a microcontroller that intercepts the current consumed by the servomotor during the gripping action and properly adjusts its aperture, with respect to the force limits suitable for each type of fruit. Experiments were performed on various objects, from elastic balls to oranges, tomatoes and sweet bell peppers. These experiments revealed that the relationship between current consumption and applied force can be accurately approximated by nonlinear expression equations and verified the good performance of the proposed force limitation technique. Consequently, there is scope for adoption by a wide range of agricultural automation systems.

1. Introduction

Given the pressing global challenges and existing limitations in agriculture, the adoption of automation in the form of agricultural robotic systems, in all facets of the agricultural industry, is seen as a viable solution. In his review study, Vougioukas [1] examines two primary issues in modern agriculture that highlight the growing interest, research and financial investment in agricultural robotics. The first issue is the urgent need to meet the demands of a growing world population with increasing purchasing power by increasing the production of high-quality food, feed and biofuel products in a sustainable way. The second is related to the shortage of available manual farm labor, which necessitates the use of agricultural robots to improve worker safety and efficiency. Furthermore, socio-economic instability and pandemics exacerbate this labor shortage [2]. The necessity to cut costs in modern farming operations and the increasing improvement of technology are additional factors contributing to the rapid developments in agricultural robotics research and application [3]. The scientific literature shows a wide range of agricultural robotics applications, mainly focused on weeding, planting, disease and insect detection, plant monitoring and phenotyping, spraying and, last but not least, harvesting [4].

In terms of harvesting, while there are well-established machine systems for harvesting crops such as wheat, soya beans and maize, the harvesting of horticultural crops such as apples, tomatoes and grapes still relies predominantly on manual labor. Bulk harvesting methods, such as tree-shaking machines, have been developed for some of these specialty crops, but most of these are only applicable to crops that are tolerant of physical strain. In this context, a significant part of the process of further automating agriculture is the development of suitable robotic grippers for handling delicate individual agricultural products. Human workers can skillfully grasp and handle objects of different shapes and sizes with minimal training, and robotic arms and grippers are designed to replicate this dexterity, allowing robots to perform similar manipulative tasks. These robotic systems can provide enhanced precision, monitoring and control, thereby improving safety and efficiency in various applications [5].

Mechanical manipulation can be defined as the act of applying force or torque to an object to induce motion or deformation, while the act of holding an object securely without slipping is referred to as grasping. Grasping and manipulation are fundamental robotic functions that also enable other diverse and more complex actions, such as locomotion and sorting, and are also capable of gathering sensory information about object properties [6]. In terms of configuration, two-finger or two-finger-claw grippers are among the most basic and widely used types of robotic grippers due to their ease of use, simplicity, availability and suitability for various applications [7]. These grippers are also relatively inexpensive to manufacture, as they are not mechanically complex. Robotic grippers are often fitted with exteroceptive sensors, such as pressure, visual, resistive and conductive sensors, to gather data about the objects they interact with and optimize their performance.

Indeed, the correct manipulation and grasping of individual fruits is of particular importance in agricultural robotics. Human workers excel at handling delicate produce with the right balance of strength and care, using their sense of touch and coordinated eye–hand movements to adapt to the task at hand, without applying excessive force or causing damage. To be effective, robotic systems must replicate these skills to ensure that the more mechanically sensitive agricultural products are handled efficiently. Compression, vibration, impact and other types of physical force can cause mechanical or physical stress to fruit and vegetables, affecting their quality and shelf life. This type of stress occurs primarily during the various stages of handling, from harvesting to transport, and can result in direct physical damage such as puncturing and deformation, which can then affect the biological processes within the produce. Bruising is one of the most common of these problems and usually appears as softened, discolored spots on the surface of the produce. Bruising is associated with lower market value and higher postharvest losses because it shortens the shelf life of produce by accelerating spoilage and reducing visual appeal [8].

A crucially important aspect of grippers used in agriculture is the way in which the forces exerted on the produce are measured, studied and used to adjust the gripper’s grip. Excessive force can easily cause damage, breakage, or deformation, while monitoring and managing gripping forces also ensures an effective grip that prevents slippage. Through the use of tactile and optical sensors, robotic grippers aim to replicate the capabilities of human hands by integrating force sensing and vision systems [9]. Many robotic grippers use force sensors to directly measure the gripping force. These sensors are often thin, flexible and capable of measuring both static and dynamic forces, while the output is typically interfaced to a microcontroller to adjust the grip based on the characteristics of the object [10]. Other solutions provide real-time feedback on the gripping force by using strain gauges positioned at different points on the gripper. Other approaches used in soft robotic grippers include flexible triboelectric sensors that have been integrated to measure both compressive and bending forces during gripping. Miniaturized force–torque sensors have also been developed for delicate tasks, such as gripping small or fragile objects. The use of fiber-optic force sensors has also gained attention, particularly in applications where high sensitivity and precision are required. Typically, the signals provided by the force sensors are processed and used to dynamically adjust the function of the robot grippers. Often embedded systems are used and a closed-loop feedback mechanism is employed, where the sensors continuously monitor the forces exerted by the gripper [11]. In any case, a good compromise between accuracy, efficiency, complexity, reliability and definitely cost should be determined, which remains an open challenge.

In this context, this study presents a simple and low-cost improvement of the ability of a servo-electric gripper to adjust its force to pick up delicate fruits without damaging them, i.e., in a non-destructive way. Specifically, this module utilizes a microcontroller that intercepts the current consumed by the servomotor during the gripping actions (via a current sensor that provides indirect feedback on the force applied) and properly adjusts its aperture, with respect to the force limits suitable for each type of fruit. Instead of the traditional force/pressure sensors, which are often prone to damage, the proposed system monitors the amperage consumption of the motor, which correlates with the resistance encountered when gripping the fruit. This method is of reduced complexity and exhibits good functionality. This control methodology has not been extensively studied in agricultural robotics. Tedford [12] first examined motor current monitoring for force estimation and identified/described a linear correlation between current consumption and gripping force. How et al. [13] further supported this concept and demonstrated its applicability with proportional–integral (PI) and proportional–derivative (PD) control algorithms. Kumar et al. [10] developed a control system combining a load-current sensor with a force sensor. Nevertheless, none of these works provided real-world performance data suitable for agricultural purposes, nor exploited the very recent technological achievements leading to compact implementations.

2. Materials and Methods

It should be mentioned that postgraduate students of agricultural engineering were involved in the development and evaluation of the smart gripper presented. They were mainly people following the Master’s program entitled “Digital Technologies and Smart Infrastructure in Agriculture”, offered by the Department of Natural Resources Management and Agricultural Engineering of the Agricultural University of Athens (AUA) in Greece. The practices applied in the above-mentioned postgraduate program are closely linked to the ongoing European Union (EU) project on digital literacy entitled “Digital agriculture for sustainable development”, with the acronym AGRITECH EU (grant agreement number: 101123258) [14].

The core of the whole system is an Arduino Uno microcontroller that controls the operation of a 3D-printed two-finger robotic gripper. The gripper uses an angular servomotor to alter its aperture. An INA219 chip-based amperage measurement unit (manufacturer Adafruit, Brooklyn, NY, USA) is intervened, in series, between the power source and the servomotor, to provide fast and real-time measurements. The INA219 sensor includes a precision amplifier, a 12-bit analog-to-digital converter (ADC) and a 0.1 ohm shunt resistor, all assembled on an integrated circuit board. The measured real-time values were communicated to the microcontroller via an I2C interface. Initially, a Tower Pro MG996R 180-degree servomotor (manufacturer Waveshare, Shenzhen, China) was utilized. The MG996R uses an internal potentiometer for feedback on the shaft’s position, allowing the built-in controller to precisely adjust the motor’s movement. It can deliver up to 11 kg/cm of torque at 6 volts and can draw up to 1400 mA of current at full load. In turn, a Feetech FT5530M servo (manufacturer Shenzhen Feetech, Shenzhen, China) replaced the MG996R, due to its higher torque and robust design, which could provide more extensive test results. The FT5330M module can deliver up to 35.5 kg/cm of torque at 7.4 V and draws up to 3900 mA of current at full load, allowing it to handle relatively heavy loads despite its small form factor. As with the servomotors, the original 3D-printed gripper was reinforced with thicker tendons, while changes were made to the gear layout to allow it to withstand greater mechanical stress without deforming or breaking. It should be noted that the power requirements of the servomotor cannot be met by the Arduino Uno’s power circuit, so an external power supply was used to avoid damage and not limit experimentation. Figure 1 provides an overview of the design of the proposed system.

In terms of software, the Arduino IDE, version 1.8.16, programming and monitoring environment was utilized for both programming and inspecting the behavior of the smart fruit gripper system. The latter task was drastically facilitated by the included Serial Monitor and Serial Plotter components, tools that preceded the meticulous data analysis being necessary to extract the relationship between the current consumed and the force being applied to the fruit by the servomotor. In order for the latter relationship to be determined, it was necessary to correlate the amperage consumption values with their corresponding force effect on typical objects such as elastic balls, oranges, tomatoes and sweet bell peppers.

For this reason, the mechanical properties of the fruits (i.e., force thresholds) were assessed by performing in parallel controlled compression tests using the Universal Texture Analyzer TA.XT2i (manufacturer Stable Micro Systems Ltd.,Godalming, United Kingdom), equipped with the 70 mm diameter compression disk. The compression speed at fruit contact was set at 0.17 mm/s as proposed by the American Society of Agricultural Engineers [15]. The key component of our approach is the use of a force sensor in lieu of the fruit under test, in such a way that the prototype smart gripper applies force to the mounted sensor instead of a fruit. The prototype dynamometer consisted of a 5 kg load cell (also called a strain gauge) and an HX711 (manufacturer AVIA Semiconductor, Xiamen, China) load cell amplifier, whose interface also needed to be connected to the Arduino Uno microcontroller. The sensor was designed to be mounted on the gripper instead of a fruit. Therefore, it was modified using 3D printing and elastic rubber materials, to match the shape and the texture of a fruit or vegetable at the contact points with the gripper. The sensor was tested for compression forces between 1 N and 30 N and was calibrated against the Universal Texture Analyzer TA.XT2i. The calibrated force sensor was used to provide accurate force measurements, i.e., to determine the relationship between the current draw of the servomotor and the force applied to the sensor.

Figure 2 depicts the prototype smart gripper, grasping a sweet bell pepper, while connected to a laptop during programming and testing. Figure 3 illustrates the force sensor being used (left part) and the concurrent amperage consumption and force measurements recorded using the Serial Plotter component of the Arduino IDE environment. The force values have been multiplied by a factor of 10 for better inspection, while the corresponding vertical axis (in N units) has been added on the far right. Current values close to the 500 mA level (horizontal line) result in forces close to 25 N. Both measured and estimated (via the calculated amperage–force relationship) force values are plotted in Figure 3 (i.e., the two lower curves).

Indeed, after collecting the amperage–force pairs, the equation expressing the force being applied upon the fruit for a specific amperage power setting of the gripper servomotor can be approximated. In fact, the numerical analysis showed that more than one type of nonlinear equation can satisfactorily describe the force–amperage relationship. Figure 4 shows typical analysis results while gripping the 500 mA target.

Efficient gripper compliance with the force thresholds, which are determined and monitored dynamically and indirectly via the amperage readings, requires the establishment of an appropriate control mechanism. This mechanism should be able to reach the desired force value comparatively quickly and maintain it (without significant fluctuations) for the time required to place the fruit, e.g., into the harvesting pallet bin. Initially, this mechanism was implemented through deterministic incremental steps up to the cut-off force value. Performing a short time averaging/smoothing of the current values before feeding them to the control algorithm was a welcome performance amelioration. The next step was to implement a dynamic PID (Proportional–Integral–Derivative) control strategy to shorten response times. Experiments showed that a control policy exploiting the integral PID component and minimizing corrections to negative errors could provide satisfactory results. Figure 5 shows an indicative code part implementing a control method variant for the embedded logic of the gripper, using the Arduino IDE development environment. For improved performance, the algorithm handles positive and negative errors asymmetrically, as different coefficients are utilized (Ki1 and Ki2, respectively). The role and the performance of the components is further clarified in Section 3.

3. Results and Discussion

While a significant part of the performance explanation graphs was intentionally included in the previous section, for a better understanding of the underlying techniques, this section provides further evaluation results. In this regard, Figure 6 shows the test instances with different fruits such as sweet bell peppers, tomatoes and oranges. Figure 7 shows the amperage consumption of the gripper for three successive deterministic gripping cycles of an elastic object. Screenshots depict the Arduino IDE Serial Plotter tool while visualizing the data arriving from the microcontroller. Amperage values were recorded for 750 mA, 500 mA and 250 mA thresholds. Similarly, Figure 8 illustrates the amperage-limiting (and thus force) limiting effect for the orange (left) and tomato (right), for the 500 mA threshold (top) and the 250 mA threshold (bottom). The 250 mA tests were performed using the improved version of the control algorithm, which has a faster response (i.e., gripper aperture adaptation) compared to the 500 mA case.

Finally, Figure 9 and Figure 10 illustrate the good limit-following behavior of the proposed system in gripping sweet bell peppers, for the 250 mA and the 500 mA targets, respectively (i.e., close to the 17 N and the 24 N force thresholds). Both estimated and actual force curves are shown. The force–amperage relationship being described had an R² metric equal to or better than 91.92, for all relevant force threshold prediction equations. For simplicity purposes, in all the above cases, the suitable force/amperage threshold values were provided via a rotary potentiometer connected to the Arduino Uno microcontroller.

The above results verify the good performance of the proposed smart fruit gripper system, despite its low cost (below 50 €) and its simplicity. This analysis also provides further insight into how automated harvesting of the fruit should be performed, depending on the type of fruit, as different fruits should be handled at different force levels. Indeed, in a more commercialized version of the whole system though, this could be performed via interaction with a small database containing thresholds per fruit type in conjunction with an automatic fruit identification technique. In this regard, a more advanced microcontroller could also analyze images in real time and identify and collect only the ripe fruits, thus improving the efficiency of the overall non-destructive process.

There is also another less apparent but equally important dimension of the work being presented. Indeed, it provides valuable paradigm on how innovative technologies in the area of electronics, software development and 3D printing can be combined together to offer practical and cost-effective solutions for agriculture, thus bridging the gap between academia and industry and equipping students with the skills needed for their careers. The latter educationally meaningful perspective signifies further investigation activities for the parties getting involved.

4. Conclusions

The objective of this research was to present and evaluate the enhancement of a microcontroller-based gripper, designed for handling delicate fruits with an emphasis on avoiding mechanical damage to the fruit. The proposed system and force control methodology uses a current sensor to monitor the amperage consumption of the gripper’s servomotor, providing indirect feedback on the force applied to the fruit, since the force applied and the current consumption have been shown to be correlated. This allows the gripper to dynamically adjust the force it applies to ensure a secure grip without causing mechanical stress to the fruit being gripped. A prototype force sensor was developed and calibrated to validate the force–current relationship, which was found to follow a polynomial model, through a regression analysis. The system was tested under controlled conditions and demonstrated its ability to maintain a predefined force threshold by regulating the position of the motor based on its current consumption. The study concludes that monitoring motor current consumption provides a reliable and cost-effective alternative to direct force/pressure sensors, simplifying the design and reducing the complexity of robotic grippers for agricultural applications. This approach and proposed methodology may offer significant potential for improving the efficiency and safety of automated harvesting systems, particularly for delicate fruits and vegetables harvested by hand.