IS-404: Automation of Harvesting Leafy Greens in indoor, hydroponic farms

The fork assembly was fabricated as a 3D printed prototype using PLA (20% infill) to evaluate structural behaviour and assembly feasibility under representative loading conditions. Initial assembly revealed a modelling oversight in the fastening design, resulting in inadequate thread engagement and joint integrity. This was addressed through the integration of heat-set inserts, which provided a robust and repeatable threaded interface, improving both structural reliability and ease of assembly without requiring re-fabrication of components.

While prior Finite Element Analysis (FEA) predicted negligible deflection of the fork members, physical testing of the printed prototype revealed observable deformation under relatively low loads. This discrepancy arises from modelling assumptions including idealised isotropic material behaviour, simplified boundary conditions, and uniform load distribution, which do not fully capture the anisotropic and layer-dependent properties of additively manufactured PLA. Additionally, the reduced stiffness associated with 20% infill, combined with geometric tolerances and minor assembly misalignments, contributed to increased compliance in the system.

The tray clamping subsystem is designed to securely fix the tray in position, ensuring stable alignment and preventing unintended motion during subsequent lowering and cutting operations.

The tray clamping mechanism employs an over-centre toggle linkage, selected for its ability to deliver a stable and repeatable clamping force that directly supports the project’s objective of improving harvesting throughput while preserving crop quality. By achieving self-locking beyond the dead-centre position, the mechanism maintains secure tray fixation without continuous actuation, enabling energy-efficient and reliable operation during batch processing and tray lowering. Compared to the alternative latch-based hook mechanism considered, the toggle configuration provides superior force consistency and positional rigidity, which is critical for ensuring precise alignment during lettuce extraction. Furthermore, its compact, planar linkage design satisfies the system’s spatial and modularity constraints, allowing seamless integration without obstructing tray intake, thereby aligning with the overall goal of a scalable, plug-and-play harvesting solution.

A hybrid prototype of the tray clamping and lowering interface was fabricated to evaluate mechanical behaviour, assembly feasibility, and load transfer characteristics under representative operating conditions. Structural components were produced using laser-cut acrylic sheets, while linkage elements utilised commercial off-the-shelf (COTS) rods for joints and pivots. Select components requiring geometric complexity were 3D printed in PLA.

Testing of the assembled prototype revealed that the clamping mechanism required the introduction of a preloading element to maintain consistent engagement. In the absence of preload, the clamp exhibited a tendency to disengage under minor disturbances, indicating insufficient internal force to sustain the over-centre locking condition. This observation directly informed the inclusion of a preload requirement in the clamping force design, ensuring reliable and repeatable operation during tray handling.

Additionally, it was observed that the single-point attachment between the clamping mechanism and the four-bar linkage resulted in localised stress concentration and reduced structural stability. The load path through a single fastener introduced undesirable rotational compliance, which could compromise alignment during the lowering process. This highlighted the need for improved load distribution through multi-point fastening or rigid joint interfaces in subsequent design iterations. For future implementations, more permanent joining methods such as welding or reinforced mounting plates are recommended to enhance structural integrity.

As identified through physical prototyping, the clamping mechanism required an internal preload to maintain consistent engagement and prevent unintended disengagement. The over-centre toggle clamp was designed to provide a total clamping force of approximately 180–200 N, ensuring stable tray fixation during lowering and extraction. This corresponds to an internal preload requirement of 15–30 N to maintain the locking condition. (Refer to Appendix E)

The concept selection matrix shows that the reverse four-bar mechanism offers the best balance of trajectory control, flexibility, and integration. The scissor lift provides comparable stability but is limited by reduced motion flexibility and bulkier packaging. The vertical linear lift enables precise motion but requires a taller, more restrictive structure, reducing feasibility. The rotaryincline concept ranked lowest, as its changing tray orientation is incompatible with stable presentation to the waterjet subsystem.

This preliminary sketch was developed to evaluate feasible tray trajectories from the entry position to the cutting plane under spatial constraints. Among the paths considered, the pink trajectory was preferred for its smooth, continuous motion with minimal directional change, while the brown trajectory offered a more compact alternative with acceptable vertical alignment.

Geometric synthesis of the reverse four-bar linkage was then carried out to generate a controlled lowering trajectory within the system envelope. The objective was not strictly vertical motion, but a smooth, constrained path that maintains tray stability and avoids interference within the 750 mm × 690 mm frame.

This flexibility was particularly important given the constraints of the design: the tray must enter from a fixed position, remain largely within a confined frame envelope (750 mm × 690 mm), and be lowered smoothly to a position suitable for root detachment, while avoiding excessive lateral drift or abrupt changes in orientation. As such, the objective was not to achieve purely vertical motion, but to synthesise a trajectory that balances smoothness, compactness, and positional control. Three candidate linkage configurations were developed and evaluated through kinematic simulation on MotionGen.io:

Three linkage configurations were evaluated using MotionGen. Iteration 1 produced looping and non-monotonic motion and was rejected. Iteration 2 generated a smooth, predominantly downward trajectory consistent with the desired motion profile. Iteration 3 achieved the required displacement in a more compact form but with increased curvature and reduced smoothness.

While Iteration 2 provided the most desirable motion characteristics from a kinematic perspective, Iteration 3 was selected for prototyping to validate structural behaviour and manufacturability under a more constrained configuration. This reflects a design approach that considers not only ideal motion performance but also practical implementation, fabrication feasibility, and system integration.

Overall, the reverse four-bar linkage was retained as the preferred mechanism due to its ability to accommodate iterative refinement and generate a spectrum of feasible trajectories without requiring fundamental changes to the system architecture. This makes it well-suited to the evolving requirements of the project, particularly as additional subsystems such as clamping and waterjet integration are incorporated.

(Refer to Appendix F for Concept Selection) Following kinematic validation of the selected linkage, the four-bar mechanism was translated into a detailed CAD model to evaluate its behaviour under realistic spatial and integration constraints. While the kinematic model established the feasibility of the desired motion trajectory, the CAD model enabled assessment of link geometry, joint placement, and physical clearances within the system.

Through this process, several key considerations emerged. Firstly, the CAD model highlighted the importance of link thickness and joint spacing, as idealised kinematic links do not account for physical interference between members during motion. Adjustments were therefore made to ensure sufficient clearance between links across the full range of motion.

Secondly, the model allowed for evaluation of packaging constraints within the frame, ensuring that the mechanism could operate without interfering with the tray, fork assembly, or surrounding structure. This was particularly critical given the requirement to maintain an unobstructed tray entry path.

To validate the kinematic behaviour of the selected four-bar mechanism beyond simulation, a scaled physical prototype was fabricated using laser-cut linkages. This prototype was assembled to replicate the motion predicted through MotionGen and CAD, allowing direct observation of the mechanism’s behaviour under real-world conditions.

The primary purpose of this prototype was to assess whether the intended trajectory of the tray (approximated by the coupler point) could be reliably achieved, and to evaluate the smoothness and feasibility of the motion across the full range of operation. Unlike purely digital models, the physical prototype enabled identification of subtle effects such as joint play, friction, and sensitivity to dimensional inaccuracies.

Through testing, it was observed that while the overall motion trend aligned with the predicted trajectory, small deviations in link lengths and pivot placement resulted in noticeable changes in the path, reinforcing the sensitivity of the mechanism to dimensional precision. This highlighted the importance of maintaining tight tolerances and careful alignment during fabrication.

Additionally, the prototype provided insight into motion smoothness and potential binding, particularly near extreme positions of the linkage. This informed subsequent refinement of link proportions and pivot locations to avoid configurations close to singularity. The exercise also demonstrated the practicality of the mechanism from an assembly perspective, allowing evaluation of joint accessibility, fastener placement, and overall constructability, which are not fully captured in simulation.

The prototyping process highlighted the broader need to prioritise design for manufacturability and assembly (DFMA). Clearances became paramount with the number of moving components in the prototype. Features that were geometrically valid in CAD required modification to improve ease of assembly, alignment, and repeatability. This included refining hole tolerances, improving access for fastening, and simplifying part interfaces to reduce assembly complexity.

On account of visible deflection of beams, alternate designs were considered and are represented as sketches:

Only the top region of the system permits the addition of structural support for beam deflection prevention; hence, two overhead reinforcement strategies were explored. The first utilises vertical extrusion members with tensioned cables to reduce fork deflection without obstructing tray entry, while the second employs an inverted V-frame to concentrate support above the fork array, improving stiffness while maintaining a clear loading path. In both approaches, cables should be thin, low-stretch, and water-resistant to ensure effective reinforcement without introducing interference or compromising system cleanliness.

Based on preliminary torque estimation (Appendix G), the tray lowering mechanism requires a motor capable of delivering approximately 4–8 Nm to reliably overcome gravitational loading, friction, and linkage inefficiencies. A low-speed, high-torque geared motor is therefore recommended to ensure controlled and repeatable motion. The motor should be placed at the input crank of the four-bar linkage, where torque requirements are minimised and motion can be directly regulated.

To identify a suitable stem separation method, four concepts were compared against the functional requirements of the system. The evaluation criteria were derived from the project objectives, with emphasis placed on cut quality, preservation of product quality, hydroponic compatibility, suitability for batch processing, maintainability, and feasibility within the project’s scope.

Among the concepts considered, waterjet cutting achieved the highest weighted score. This was primarily due to its non-contact cutting action, reduced risk of tissue compression, compatibility with wet hydroponic environments, and strong potential for repeatable batch operation. By contrast, laser cutting was disadvantaged by thermal damage risk, mechanical blades by wear and contact-induced bruising, and rope-based cutting by inconsistent separation and contamination concerns.

Waterjet cutting preserves lettuce quality and extends shelf life due to its non-contact, low-damage cutting mechanism. Unlike mechanical blades, which introduce compressive forces and tissue tearing, waterjets minimise cellular disruption at the cut interface. This reduces the release of intracellular fluids that accelerate enzymatic browning and microbial growth. Additionally, the waterjet process can help remove damaged cells and surface contaminants during cutting, resulting in lower initial microbial loads. As a result, waterjet-cut lettuce exhibits reduced discoloration and slower degradation, contributing to improved freshness and extended shelf life compared to conventional cutting methods (Hashish), 2025). (Refer to Appendix H for comparative analysis)

This consideration is further reinforced by stakeholder feedback, which highlighted that while the concept is promising, there is a risk of inadvertently damaging adjacent crops during operation. This underscores the need for precise targeting and carefully controlled cutting parameters, particularly within the constraints of a densely packed tray configuration.

This risk is mitigated through the use of a localized, high-precision waterjet, where the jet diameter and pressure are tuned to act only at the stem interface.

A first-pass theoretical sizing calculation was performed to identify a low-pressure waterjet regime capable of separating hydroponically grown lettuce stems. The jet was modelled using Bernoulli-based nozzle flow and momentum transfer relations, and the resulting jet force was converted into an equivalent stress acting on the stem cross-section. By requiring the jet-induced stress to exceed the estimated stem failure stress with a modest safety factor, the minimum pressure requirement was expressed as a function of nozzle diameter, stem diameter, and discharge coefficient. For a representative stem diameter of 10 mm, a nozzle diameter of 0.5 mm, a discharge coefficient of 0.9, and an assumed tissue failure stress of 0.20 MPa, the required pressure was estimated to be approximately 74 MPa. (Refer to Appendix I for calculations)

This places the system within a low-pressure regime relative to conventional industrial waterjet applications, while remaining theoretically sufficient for soft biological material. The calculation is intended as an initial sizing estimate and would require refinement through material testing and controlled validation.

To validate the feasibility of the selected concept, industry consultation was conducted with FedJet, a company specialising in waterjet technologies. Through this engagement, preliminary testing was performed on lettuce stems using a waterjet system, demonstrating that clean and effective separation is achievable under controlled conditions. These findings provide practical confirmation that waterjet-based cutting is a viable approach for harvesting lettuce, supporting its selection for further development in this project.

The dashed lines (A–C) indicate the different cutting heights tested by FedJet to evaluate optimal stem separation points for hydroponic lettuce.

While preliminary validation confirms that waterjet-based cutting is technically feasible for hydroponic lettuce, full-scale experimental testing was not conducted within the scope of this project due to financial constraints. Commercial waterjet systems capable of operating within the required pressure range represent a significant capital investment, with quoted system costs on the order of USD 9,700 (EXW) based on direct correspondence with FedJet.

From a system integration perspective, several design considerations must be addressed for effective implementation of the waterjet subsystem. Firstly, water accumulation within the tray during cutting should be accounted for, as splashing and jet rebound may introduce additional transient loading; this suggests incorporating drainage pathways and designing for slightly higher effective loads. Secondly, the surface positioned behind the cutting zone must be designed to safely absorb or dissipate jet energy without causing secondary splashing or damage, potentially through compliant or sacrificial backing materials. Thirdly, effective water management is necessary, including controlled drainage, collection, and possible recirculation, to prevent pooling and ensure consistent operation within a wet hydroponic environment. Finally, preliminary consultation indicates that a simple microcontroller-based system (e.g., Arduino) may be sufficient for basic actuation control and sensing; however, requirements for precision, feedback integration, and robustness under industrial conditions would need to be validated. A thorough further investigation is required.

Firstly, the current tray-support structure relies on single-point fastening, which is insufficient for maintaining stability under load. This introduces the risk of localised rotation and uneven load distribution, indicating the need for multi-point constraint or improved joint design to ensure structural rigidity.

Secondly, the alternative designs to prevent beam deflection of the fork members have been proposed but are yet to be designed and tested. While initial simulations suggested acceptable deformation levels, physical prototyping demonstrated noticeable deflection, particularly under realistic loading conditions.

From a systems perspective, the absence of sensing and feedback components limits the ability of the system to detect tray presence, alignment, and successful engagement. This reduces robustness and introduces potential failure modes during operation.

Finally, environmental considerations such as water exposure from the waterjet process have not been fully addressed. The current design lacks protective features, such as rubber sealing or splash guards, which are necessary to prevent damage to structural and mechanical components.

Future development of the system will focus on improving geometric adaptability, structural performance, and integration into a fully automated harvesting workflow.

Firstly, the current design is based on a fixed tray geometry. Future iterations should incorporate adjustable or modular fork configurations to accommodate variations in tray dimensions and crop types. This would improve compatibility across different vertical farming systems and reduce reliance on a single standardised tray design.

Secondly, the structural members supporting the tray may be refined to better match the natural geometry of the lettuce. Replacing the current rectangular extrusion bars with inverted triangular profiles could improve clearance around the plant base while maintaining sufficient structural stiffness, thereby reducing the risk of interference during handling.

A key extension of the system is the integration of a commercially available waterjet cutting unit to enable full automation of the root detachment process. This will require alignment between tray positioning and cutting path, as well as consideration of enclosure design, splash management, and operational safety. Successful integration would transition the system from a handling prototype to a complete harvesting solution.

From a systems perspective, the addition of sensing and control elements is necessary to improve reliability. This includes detecting tray presence, verifying alignment, and confirming successful clamping, enabling more robust and repeatable operation.

Further work is also required to address structural performance, particularly the deflection observed in the fork and linkage components. This may involve optimisation of cross-sectional geometry, material selection, or support conditions to ensure consistent alignment under load.

Finally, opportunities exist to improve overall system efficiency by leveraging existing infrastructure within indoor vertical farms. In particular, the feasibility of utilising excess energy from LED lighting systems to power auxiliary components could be explored, reducing additional energy demand.

This work validates the feasibility of a modular, mechanically driven harvesting system for hydroponic lettuce through systematic design, kinematic synthesis, and physical prototyping. The study highlights critical interactions between structural behaviour, motion control, and system integration, demonstrating the importance of iterative validation in translating theoretical models into functional designs. Collectively, the outcomes provide a robust engineering foundation for scalable harvesting automation in indoor vertical farming systems.

Béguin, P., et al. (2004). Effect of cutting method on quality and shelf life of fresh-cut lettuce. Journal of Food Science, 69(6), FMS227–FMS232.

Cantwell, M., Nie, X., & Hong, G. (2006). Impact of waterjet cutting on quality of fresh-cut lettuce. Postharvest Biology and Technology, 40(2), 152–160. https://doi.org/10.1016/j.postharvbio.2005.12.012

Economist Impact. (2022). Global food security index 2022. The Economist Group.

Hashish, M. (2025). Food processing with ultra-high-pressure waterjets. Applied Sciences, 15(11), 6246. https://doi.org/10.3390/app15116246

Intergovernmental Panel on Climate Change. (2022). Climate change 2022: Impacts, adaptation and vulnerability. Cambridge University Press.

Laborde, D., Martin, W., Swinnen, J., & Vos, R. (2020). COVID-19 risks to global food security. Science, 369(6503), 500–502. https://doi.org/10.1126/science.abc4765

Low, Y. (2026, January 7). Singapore food output gets S$80 million boost with opening of Greenphyto’s vertical farm, tallest in the world. The Business Times.

Reuters. (2020, March 17). Singapore PM says food supply to continue from Malaysia despite movement restrictions. Reuters.

Schield, M., & Harriott, B. L. (1973). Cutting lettuce stems with a water jet. Transactions of the ASAE, 16(3), 440–442. https://doi.org/10.13031/2013.37537

Singapore Food Agency. (2023). Singapore food statistics 2022. https://www.sfa.gov.sg

Singapore Food Agency. (2025). Singapore food statistics 2024. https://www.sfa.gov.sg

Singapore Food Agency. (2025, November 4). Speech by Ms Grace Fu, Minister for Sustainability and the Environment, at the Asia-Pacific Agri-Food Innovation Summit. https://www.sfa.gov.sg

Singapore Food Agency. (2026). Singapore continues to strengthen its food supply resilience. https://www.sfa.gov.sg

The data presented in Appendix A highlights that Singapore’s food system vulnerability is structurally driven by limited domestic production capacity and exposure to external risks. Lower scores in dimensions such as disaster risk management and environmental resilience reflect the system’s reduced ability to absorb and recover from disruptions, particularly given its heavy reliance on imported food. Additionally, weaker performance in agricultural availability metrics indicates constrained local production capacity, reinforcing the lack of buffering against supply shocks. Together, these results show that while the system is functional under stable conditions, it remains inherently exposed under disruption. This directly supports the motivation of the present project, which aims to improve local production throughput through automation, thereby strengthening resilience within these constraints.

Option A outperforms the alternatives under the selection criteria. It protects the plant by supporting it entirely from below, avoiding hold-induced leaf damage, and it leverages the existing tray geometry without requiring modifications to the vertical-farm workspace. Its static forks and simple lift motion keep the mechanism mechanically minimal, with few moving parts, low failure probability, and essentially zero energy demand for this function.

Option B is unsuitable because its sliding blades require precise stem alignment, create sap-collecting crevices that demand frequent cleaning, and present upward edges that can snag or cut lower leaves, making hygiene and leaf safety difficult to maintain.

Option C, though similar, introduces dual sliding arms that increase mechanical and control complexity.

Option D’s compliant behaviour is appealing, but relying on an external robotic arm, valves, and a compressor makes it the most infrastructure-heavy, energy-demanding, and least retrofit-friendly option. Although it was the original approach, it proved misaligned with the constraints of this application.

To ensure dimensional accuracy, the tray geometry was physically measured at the NUS Centre for Regenerative Agriculture, where representative hydroponic trays used in indoor farming systems were accessed.

The clamping mechanism is a passive over-centre toggle and does not rely on a dedicated actuator. The required holding capacity was therefore estimated from friction rather than actuator thrust. The loaded tray mass was taken as 3.85 kg, corresponding to a static weight of 37.8 N. To account for startup effects, guide friction, and root drag during downward extraction, the pull load was conservatively taken as 1.5–2.0 times the tray weight, i.e. 56.7–75.5 N. Assuming a friction coefficient of 0.4–0.5 between the tray and clamp pads, the required total normal clamping force is approximately 113–189 N. A design value of 180–200 N total clamping force was therefore adopted. Since the clamp operates as an over-centre toggle, the internal linkage force needed to maintain lock is much smaller than the normal clamping force. For a lock angle of approximately 5–10° from dead centre, an internal preload of about 15–30 N is sufficient to generate the required clamping force.

The selection criteria for the tray lowering mechanism were defined based on their direct relevance to system performance, integration, and feasibility within the project scope. Tray stability, particularly maintaining a horizontal orientation during operation, is critical to ensure consistent interaction between the waterjet and the lettuce roots. Any unintended tilt could result in uneven cutting and increase the risk of crop damage. Compatibility with a conveyor-fed input is also a key requirement, as the mechanism must integrate seamlessly with a front-fed tray system without obstructing entry, making this a strict architectural constraint.

Control over the motion trajectory is equally important, as the lowering action must be smooth and predictable. This ensures that the lettuce placement is not disturbed, a consistent cutting height is maintained, and dynamic loads on the system are minimised. In addition, suitability for batch processing is essential, since the system is expected to operate repeatedly with minimal recalibration. High repeatability directly impacts throughput and overall system efficiency.

Design flexibility was also considered, given the evolving nature of constraints such as waterjet integration, frame dimensions, and potential variations in tray design. The mechanism should therefore allow for adjustments in link lengths, motion paths, and facilitate rapid iteration during CAD modelling and prototyping. Complexity, in terms of design difficulty, captures the challenges associated with kinematic synthesis, analysis, optimisation, and control requirements. Finally, buildability evaluates the practical feasibility of manufacturing the mechanism, including ease of fabrication using methods such as laser cutting, 3D printing, and off-the-shelf components, as well as sensitivity to assembly tolerances and alignment requirements.

The scissor lift demonstrates strong performance due to its inherently stable vertical motion, which maintains tray orientation effectively and ensures high repeatability. It integrates well beneath the tray without obstructing entry and is simple to design and fabricate using basic linkages. However, its primary limitation lies in its lack of design flexibility, as the motion trajectory is largely fixed and cannot be easily tuned beyond simple scaling.

The 4-bar linkage, which was ultimately selected, offers the best overall balance of performance and adaptability. While slight orientation deviations may occur depending on geometry, it excels in compatibility with the front-fed system and provides fully tunable motion through link length optimisation. This allows precise control over the trajectory while maintaining high repeatability. Its key advantage lies in its design flexibility, enabling adaptation to constraints such as waterjet integration and frame limitations. Although it introduces moderate complexity in terms of kinematic synthesis and requires tighter alignment during fabrication, it remains feasible to build. Overall, the 4-bar linkage stands out because it achieves trajectory tunability without compromising repeatability, making it particularly well-suited to the system’s functional requirements.

In contrast, the rotary or incline mechanism was deemed unsuitable due to its fundamental mismatch with system needs. Its tendency to introduce tray tilt makes it incompatible with the requirement for stable, horizontal cutting. Additionally, it offers poor control over vertical motion, lower repeatability, and potential interference with tray entry depending on its configuration. While relatively simple to construct, these limitations outweigh its benefits, leading to its elimination from consideration.

This motor was inherited from a previous project and provided as part of the existing system hardware.

Laser cutting offers a non-contact approach and precise energy delivery; however, it introduces thermal effects that are unsuitable for fresh produce. The high moisture content of lettuce leads to localised heating, causing tissue degradation, dehydration, and potential browning at the cut interface. Additionally, laser systems require optical alignment, enclosure, and safety considerations, reducing feasibility within a compact, wet farming environment.

Mechanical blades are simple, compact, and widely used in agricultural applications. However, as a contact-based method, they introduce compressive forces prior to cutting, which can result in bruising and inconsistent separation. Blade wear and dulling over repeated cycles further reduce cut quality and require frequent maintenance. While feasible and easy to implement, performance variability and potential impact on product quality limit its suitability. Lettuce bleeding is also noted from experiential harvesting exercise. Sap gets secreted which affects blade longevity and questions hygiene standards.

Waterjet cutting provides a non-contact cutting mechanism that minimises mechanical stress on the crop. This reduces tissue compression and enables cleaner separation compared to blade-based methods. Its compatibility with wet, hydroponic environments further supports integration within indoor vertical farming systems. Additionally, the absence of blade wear improves consistency and reduces maintenance requirements. While the system introduces complexity in terms of pressurisation and fluid management, it offers the best balance between cut quality, repeatability, and system compatibility.

Rope based cutting offers a mechanically simple approach; however, it relies on frictional contact and tension, which can result in tearing rather than clean cutting. The method also presents a risk of material degradation and particulate contamination (microplastics), particularly if polymer filaments are used. Furthermore, maintaining consistent tension and alignment across multiple cuts is challenging, reducing reliability in batch processing scenarios.