**3. Results and Discussion**

After performing the hypothesis test, we estimated that there were statistically significant differences between the grafting times used for the different tested work rates (Table 1).

For each velocity tested (between TS1 and TS6), the variability of time used in each test unit was mainly due to singularities that facilitated the manipulation and the development of the grafting work, to a greater or lesser extent. Singularities included the position of the alveolus in each row of work, the natural variability of the seedling emergence point within each alveolus, and the unique growth morphology of each seedling, among others. At low velocities, the difference for each work velocity was clear, given that the time taken to solve these singularities was less significant compared to the time spent in tasks not affected by these singularities. However, at high speeds, these factors became increasingly important and, to some extent, determined the time spent in each test unit.

Grouping the data by test speed, analyzing the failures associated with each control point that were recorded for the different velocities, and performing the hypothesis test, we estimated that, based on the Tukey's tests, there were statistically significant differences between the groups of different assay speeds (Table 2).

At low test rates, the success rate was higher, greater than 90% for speeds equal to or lower than TS3, and there was a significantly increasing graft failure for operating speeds equal to or greater than TS4. Low speeds, between TS1 and TS3, had similar behavior in terms of failures; therefore, we consider that their differences were derived from chance and not from the working speed itself. Therefore, it is clear that the TS3 production speeds are more attractive, given that they had a higher production ratio associated with a low failure rate.

The relationship linking the number of grafts/hour for each of the tested rates (TS1 to TS6) can be considered practically linear and was only altered by the random parameters derived, to a great extent, from the natural variability of the work seedling growth. Such nuances of correction barely affected the speed, which was largely marked by that established for both robots for each test, although the working speed of the auxiliary devices was kept constant. In addition, the percentage of successes/failures associated with each of the test rates (TS1 to TS6), considering their evolution, allowed us to observe a behavior similar to a quadratic function (Figure 9).

**Table 1.** Comparison between the mean times spent in grafting task for the six tested speeds. Grouping of comparisons applying Tukey's HSD test (honestly significant difference). Significance level *p* < 0.05.


The different letters show significative differences.

**Table 2.** Comparison between the means for grafting failures for the six speeds tested. Grouping of comparisons applying Tukey's HSD test (honestly significant difference). Significance level *p* < 0.05.


The different letters show significative differences.

**Figure 9.** Grafts/hour versus grafting success for different test speeds (TS1 to TS6). Success rate includes the typical error. "Success before healing chamber" has been included to evaluate the influence on the global success of the grafting process.

At the working speeds of the TS3 and TS4 robots (210 and 240 grafts/hour, respectively), the number of grafts estimated for manual expert workers was already surpassed, i.e., a maximum of 150–240 grafts/hour [7,15], and an average of approximately no more than 1000 grafts per person per day [30,33,34]. In addition, the success rates for manual grafting are not usually very high, between 81% and 91% [35]. In part, this rate of failure may come from long hours in a hostile work environment, characterized by high humidity and temperature. The success rates achieved by the robotic graft were quite similar to those achieved by manual workers, reaching 90.0% for the TS3 speed of 300 mm/s and 85.3% for the TS4 speed of 400 mm/s.

For large cutting angles in splicing grafts, the difference between the diameters of the linked stems became less important as long as they stayed within certain a range [31]. For cuts at a 60◦ bevel, failure in the healing and grafting of the middle graft was between 3% and 6% of failures [27]. The difference between the number of successes considered after grafting and the number of successes recorded after the healing period in the chamber confirmed this parameter (less than 4%), and except for errors not visually detected in the grafting process, we associated the percentage of losses in the chamber to random problems in the healing process itself.

The established isolated intermediate control points allowed us to record the cause for each failed graft. Studying the origin of the graft failures associated with each velocity in detail, we determined that at low test rates, the failures detected in the grafting process in the GE phase (grip and extraction) were slightly more significant. This result may be because the seedling, when extracted from its alveolus at a lower velocity, experiences lower extraction acceleration, with a consequently greater resistance and grip to the alveolus walls. As a result, the roots tend to remain adhered and occasionally incur damage and tears; traction damage to the stem or alterations to the other plants in the tray may also occur, among other problems.

However, it was observed that, as the test speed increased, the number of failures in this phase decreased, as did the number of failures in the operations of adaptation and placement of the seedlings, such as the AC (approach the cutting zone) and ACD (approach the clip dispensing zone) phases, and in the tasks properly conducted by these tools: C (cutting process) and CPP (clip preparation and placement). In part, this was due to seedling management from certain working speeds, where there was a substantial acceleration in the displacements between points, and with it, the inertia experienced on the seedlings, which, together with their root ball (semi-compressed coconut fibre substrate) could suffer greater damage and tearing when experiencing such sudden changes of state. In addition, it was observed that high accelerations led to excessive seedling balance by the ends not held by the robot clamp (root ball and stem), causing the robot to lose its reference point at rest or to not return to it in time, thus spoiling the graft. The development of the grafting process is shown in Figure 10.

As the working speed increased, the times spent in the tasks performed by the external working tools became more important compared to the times spent in the displacement and pre-positioning of the seedlings in front of the tools. This factor is due to the fixed value of the velocities of the pneumatic devices in response to an incremental increase in the robot velocities (Figure 11).

At speeds equal to or lower than TS3, the recorded success rate was relatively good at approximately 90%, but it decreased substantially at higher speeds. The next tested speed, TS4, had a significantly different percentage of recorded failure, five points lower or 85.3%. It is important to assess the success associated with each work rate, because it is the factor that makes it feasible as an alternative to manual grafting, because the system evaluated is scalable in terms of systems and tools operating in parallel. That is, the clamp or end element could be adapted by cloning two or more gripping systems in parallel for the seedlings, and, to the same extent, the auxiliary devices or tools acting on the seedlings could be cloned, thus multiplying the number of plants grafted per hour, maintaining similar success rates. In addition, regarding these working speeds (TS3 and TS4), the number of grafts capable of being developed manually began to be exceeded. Therefore, when evaluating the grafts/success ratio, we estimated a better average behavior for velocities close to TS3 (210 grafts/hour).

**Figure 10.** Success associated with each check point for the different test speeds, reflected in a percentage of successful plants in the grafting process. Check Points: approach the input trays (AIT), grip and extraction (GE), approach the cutting zone (AC), cutting process (C), approach the clip dispensing zone (ACD), clip preparation and placement (CPP), approach the output trays (AOT), insertion on trays and release of plants (ITR), return home (RH), and post-graft losses (PGL).

**Figure 11.** Average time spent in the development of each phase between control points. Check Points: approach the input trays (AIT), grip and extraction (GE), approach the cutting zone (AC), cutting process (C), approach the clip dispensing zone (ACD), clip preparation and placement (CPP), approach the output trays (AOT), insertion on trays and release of plants (ITR), and return home (RH).

Numerous studies have collected the test results of prototypes and commercial devices for the automated grafting of horticultural seedlings over the last three decades [3,5,7,23,36–44], and the results for dozens of pieces of equipment have been collected, mainly from Southeast Asia (Japan, Korea, and China mainly) and Europe.

Comparatively, we can refer to four factors that determine the convenience of and interest in the study equipment compared to other existing equipment: (a) the flexibility in terms of the horticultural family of work and the grafting method developed; (b) the degree of automation and the number of operators involved in the process; (c) system velocity (grafts/hour) and system efficiency; and (d) the price of the equipment.

(a) There are devices prepared exclusively for use with Solanaceae and others that allow working with Solanaceae and Cucurbitaceae. These pieces of equipment are mostly characterized by being inflexible. The study equipment, based on industrial robotics supported by low-cost external equipment, is specifically intended for splicing grafts, but enables easy and economic reconversion and readaptability, as it is able to work with other horticultural families and to apply other grafting methods. It even has the ability to perform other tasks in the seedbeds or productive environments.

(b) The existing prototypes and commercial devices present a different degree of automation, ranging from simple tools to help with grafting (cutting or dispensing the clip), to semi-automated equipment that requires the participation of two to three operators, to fully equipped devices. These are all automated and can be managed by a single operator. Nevertheless, they claim a high homogeneity in the work seedlings, requiring the important tasks of pre-sorting and pairing between seedlings, which to some extent tarnishes the autonomy of the system. The study equipment, by working with a high cutting angle, allowed us to avoid these previous tasks of searching and matching between the diameters of the workplaces and therefore enjoys a high degree of autonomy.

(c) Regarding the number of grafts per hour, the majority of studies present equipment ranging between 200 and 1200 grafts/hour. These values are much higher than those achieved using this equipment (210–240 grafts/hour), but these results are sufficient compared to manual grafting, and improvement is feasible by replicating terminal systems, allowing for working in parallel and over several plants simultaneously. In contrast, the system has a high success rate or efficiency of approximately 90%.

(d) The current price trend of industrial robotics, together with the use of simple, low-cost auxiliary equipment, allows us to estimate that the studied system is a rapidly amortizable system, with an initial investment in robots no greater than the biannual cost associated with the minimum interprofessional wage in countries [25], and an additional base cost of auxiliary and control equipment which is not higher than other interprofessional minimum wages (MW). This implies a total base investment of 5 MW (~€60,000). Faced with this, the estimated costs of high automation systems robotic equipment in high productivity environments (100 million plants per year) are estimated at investments above ~\$7,500,000 [45]. Comparatively, and estimating our system working at approximately 230 grafts/hour, we would yield approximately 2 million grafts per year per piece of equipment, which would imply approximately 50 systems working in parallel to reach 100 million grafts. Such scaling would involve an investment of ~€3,000,000, well below the investment necessary for other robotic equipment.

#### **4. Conclusions**

The splicing technique, widely used for Solanaceae such as tomato, has the advantage of being simple and methodical, and therefore easily automatable. In the last three decades, there have been numerous attempts to develop equipment to deal with the automated grafting of horticultural plants, and the challenge of using conventional industrial robotics to perform the splicing graft process can provide a great opportunity. The robotic cell tested herein is based on two industrial robots and low-cost passive auxiliary units.

The use of low speeds, between 100 mm/s and 300 mm/s, allows ratios close to a 90% success rate to be maintained. At medium–high velocities, between 400 mm/s and 500 mm/s, success ratios were still acceptable above 80%. However, at a test speed of 600 mm/s, there was a considerable decrease in the success rate to less than 70%.

Consequently, we conclude that it is advisable to use a velocity close to 300 mm/s (90.0% success), which allows working at speeds higher than those estimated for manual expert workers, approximately 150–240 grafts/hour (with success rates between 81% and 91%). Decreasing the work rate below that point did not substantially improve the success rate.

**Author Contributions:** J.-L.P.-A. conceived and designed the experiments, performed the experiments, collected and analyzed the data, interpreted the results and developed the manuscript. Á.C.-O. conceived and designed the experiments, analyzed the data, interpreted the results and developed the manuscript. C.-C.M.-G. conceived, designed and performed the experiments. I.G. and M.G.G. provided constructive suggestions on experiment analysis.

**Funding:** This research received no external funding.

**Acknowledgments:** This study was supported by Tecnova, Technological Center: Foundation for Auxiliary Technologies for Agriculture. The authors would like to acknowledge all the employees involved for their contributions to the experimental setting and data collection.

**Conflicts of Interest:** The authors declare no conflicts of interest.
