Height-Based Stratification in Greenhouse Harvesting Robotics: A Review from Ground-Level to High-Wire Crops

Logical Operator Operator

Search Text

Search Type

add_circle_outline

remove_circle_outline

Article Menu

Font Type:

Arial Georgia Verdana

Font Size:

Aa Aa Aa

Line Spacing:

  

Column Width:

  

Background:

Open Access Review

by
Zuhui Zhou

1,†,
Yile Chen

1,†,
Wenshuo Gao

1,*,
Xinpeng Wang

2,
Xuan Liang

1 and
Xifeng Liang

1

1

School of Mechanical and Electrical Engineering, China Jiliang University, Hangzhou 310018, China

2

Graduate School of Chinese Academy of Agricultural Sciences, Beijing 100081, China

Author to whom correspondence should be addressed.

†

These authors contributed equally to this work.

Submission received: 11 June 2026 / Revised: 23 July 2026 / Accepted: 30 July 2026 / Published: 11 August 2026

Abstract

Labor shortages and the push for higher greenhouse efficiency have accelerated interest in automated harvesting. However, the development of a universal harvesting robot has been constrained by large variations in crop architecture, especially plant height. In this review, a height-based stratification of greenhouse harvesting robots and transferable high-wire crop harvesters is presented, covering ground-level crops (<0.6 m, e.g., strawberry), medium-height crops (0.6–1.5 m, e.g., tomato), and high-wire crops (>1.5 m, e.g., trellised cucumber). For each height layer, key design features, technical progress, prototype performance, and common obstacles—including fruit occlusion, mechanical crop damage, unreliable operation, and high commercial costs—are analyzed. Future efforts should target intelligent perception, soft end-effectors, and height-specific solutions (swarm robotics for ground crops, modular hybrid designs for medium crops, infrastructure co-design for high-wire crops). By using plant height as the primary stratification criterion, a design-oriented framework is provided, distinct from conventional crop-type or mechanism-based categorizations.

1. Introduction

The advancement of protected agriculture, particularly greenhouse cultivation, has become increasingly pivotal for ensuring year-round food security and enhancing production efficiency amidst global population growth [1]. Greenhouses provide a controlled environment that mitigates the adverse effects of climate change and seasonal fluctuations, thus offering a stable platform for high-density, high-quality crop production. As a representative example of this global agricultural trend, China is modernizing its agricultural sector through the mechanization and intelligent upgrading of production equipment, a development aligned with the global transition toward “Agriculture 4.0” [2].

Within this framework, automated harvesting has emerged as a core technology for mitigating the challenges of escalating labor costs and seasonal workforce shortages. These challenges are particularly acute in labor-intensive greenhouse operations [3]. Harsh working conditions—high humidity, repetitive motion, and elevated temperatures—combined with relatively low wages have made manual harvesting increasingly unpopular among the younger generation. Consequently, advanced robotic interventions are required to sustain agricultural viability [4,5]. The harvesting manipulator and its critical end-effector for picking and handling are considered the functional heart of any automated harvesting system [6]. The harvesting manipulator is defined as the mechanical arm responsible for motion execution. The end-effector refers to the terminal actuator directly contacting fruits. It serves as the primary interface between mechatronic systems and delicate biological fruit.

In recent years, much research has focused on soft robotic grippers that mimic human finger compliance. However, the remarkable architectural diversity of greenhouse crops remains a formidable obstacle to universal manipulators [7,8]. Unlike industrial environments, agricultural settings present unstructured challenges: fruit size, shape, and ripeness often vary even within the same cultivar. Consequently, the performance of harvesting robots—measured by success rate, cycle time, and damage rate—frequently falls short of commercial requirements. A harvesting robot refers to the complete system, including mobile platform, manipulator, vision system, and control system. Success rates vary significantly and drop sharply when robots encounter complex morphologies or unpredictable conditions such as fluctuating light and overlapping foliage [9,10].

Among various morphological traits, crop growth height is a primary factor dictating manipulator design. The vertical distribution of fruit directly determines the required workspace, kinematic configuration (e.g., degrees of freedom), and structural stiffness needed for precision at different reaches [11,12]. Furthermore, growth height influences perception complexity. For example, ground-level crops often need top-down sensing to navigate soil occlusion, whereas high-wire crops require side-view vision systems with obstacle-avoidance algorithms [13].

Despite its critical importance, crop height has not been systematically adopted as a classification criterion in existing reviews of harvesting manipulators. Current surveys typically fall into two categories: crop-based classifications, which organize technologies by individual species and are intuitive for agronomic research but often redundantly describe structurally similar manipulators across different crops; and mechanism-based classifications, which group technologies by hardware form and clarify mechanical distinctions but remain decoupled from practical agronomic constraints and field application requirements [14,15]. Neither approach prioritizes height as a primary design driver, leaving a gap between hardware taxonomies and the crop-driven spatial constraints that fundamentally shape manipulator architecture.

This gap is significant because a robot’s operational efficiency depends on how well its manipulator adapts to specific crop growth patterns and height distributions [16]. A mismatch between workspace and fruit spatial distribution leads to “dead zones” and reduced coverage.

To address this gap, a literature search was conducted in Web of Science, Scopus, and IEEE Xplore using keywords “greenhouse harvesting robot”, “harvesting manipulator”, “fruit picking”, “crop height”, and “agricultural robot”. The search covered papers published between 2010 and 2026, up to May 2026. Only peer-reviewed journal articles and conference proceedings describing physical prototypes or validated simulation studies of harvesting manipulators for greenhouse crops were included. Duplicate records were identified and removed using reference management software. “Reports not retrieved” refers to articles for which the full text could not be obtained through the Web of Science, Scopus, or IEEE Xplore databases, whereas “Full text unavailable” refers to conference abstracts without full manuscript publications. From the retrieved literature, a keyword co-occurrence network was derived using VOS viewer, as shown in Figure 1.

Figure 1. VOS viewer keyword co-occurrence network.

Following the PRISMA 2020 guidelines, a total of 88 papers were selected from the initial 342 for detailed analysis based on relevance to height-dependent design. The inclusion and exclusion criteria are listed in Table 1, and the PRISMA flow diagram is presented in Figure 2. Relevance was defined as papers that explicitly reported manipulator design parameters related to crop height or presented experimental results for a specific height layer, i.e., ground-level, medium-height, or high-wire. In addition, 16 references were cited as background material but were not included in the systematic review. The literature screening was conducted independently by two reviewers. The inter-rater agreement was substantial, with a Cohen’s κ of approximately 0.85. Any disagreements regarding paper inclusion were resolved through discussion with a third senior researcher to reach a consensus.

Figure 2. PRISMA 2020 flow diagram. Note: Five records were automatically excluded by the database search engines during the initial retrieval process, as they did not match the predefined search strings and were deemed irrelevant to the review topic.

Table 1. Inclusion and exclusion criteria (PRISMA 2020).

Based on this curated corpus, plant height is adopted as the primary classification criterion. A height-based stratification of greenhouse harvesting manipulators is proposed, spanning from ground-level crops such as strawberries to high-wire crops such as trellised cucumber. This review focuses on three height layers, namely ground-level, medium-height, and high-wire. Below-soil crops are excluded from this classification because they are rarely harvested by dedicated greenhouse robots due to different working principles and perception challenges; a brief discussion is provided in Section 2.4. To substantiate the proposed classification, a comparative analysis of the primary design specifications imposed by each height layer is presented in Figure 3. The analysis reveals that growth height does not merely influence a single parameter but systematically constrains the entire manipulator architecture, from kinematic configuration to perception strategy. For example, perception algorithms developed for occlusion handling in medium-height crops may be adapted to high-wire crops after suitable retraining. By using height as an integrative variable that encapsulates multiple design-critical factors, a more holistic framework is provided for the design and evaluation of greenhouse harvesting manipulators. This framework highlights both cross-layer opportunities and barriers to technology transfer.

Figure 3. The height distribution of the greenhouse crops’ fruit.

2. State of the Art in Height-Based Stratification of Harvesting Manipulators

Greenhouse harvesting manipulators are reviewed here according to crop growth height. Three height layers are defined, based on the classification established in the abstract and Table 2: ground-level crops (<0.6 m, e.g., strawberry), medium-height crops (0.6–1.5 m, e.g., tomato and cucumber), and high-wire crops (>1.5 m, e.g., trellised grapes). The classification logic is directly linked to the technical requirements imposed by each layer. The parameter ranges in Table 2 are derived from a qualitative synthesis of the reviewed literature to illustrate general design trends, not as precise specifications. Payload typically includes the end-effector and fruit, positioning accuracy refers to static repeatability as reported in the original studies, generally excluding mobile-platform-induced errors unless explicitly stated.

Table 2. Comparative analysis of design requirements across height categories.

These specific height thresholds (0.6 m and 1.5 m) are proposed based on a systematic synthesis of the crop height distributions reported in the 88 reviewed studies, combined with standard greenhouse cultivation practices and typical kinematic workspace limitations of agricultural manipulators. The 0.6 m boundary corresponds to the maximum canopy height of ground-level crops such as strawberries in ridge or table-top cultivation systems, above which fruits are no longer ground-accessible and require different manipulator configurations. The 1.5 m boundary separates medium-height crops from high-wire crops, corresponding to the practical reach limit of conventional multi-purpose manipulators; crops exceeding this height typically require gantry-mounted or long-reach arms with extended workspaces. These thresholds are therefore presented as a research-derived classification framework rather than a citation of any single external standard.

As crop height increases, the manipulator must contend with a progressive shift in working zone orientation (from downward to overhead), an expanding reach requirement that amplifies positioning errors through joint compliance [11,12], and a transition in perception complexity (from ground-level to canopy-penetrating 3D localization) [13]. These systematic trends validate growth height as a functionally meaningful axis for organizing manipulator technologies.

The specific design imperatives for each height layer are as follows:

Ground-level crops (e.g., strawberries): Manipulators with a small working radius, high-precision positioning, and damage-resistant end-effectors are needed. These requirements are driven by compact growth (dense planting, narrow aisles) and fragile fruits [17,18,19,20].

Medium-height crops (e.g., tomatoes, cucumbers): Manipulators that balance flexibility, load capacity, and crop-specific handling (e.g., gentle grasping for tomatoes) are required. This is attributed to varied fruit positions, diverse morphologies, and moderate occlusion [21,22,23].

High-wire crops (e.g., trellised grapes): Manipulators with extended working ranges, stable high-altitude performance, and branch or trellis navigation capabilities are required. These features address long-reach needs, complex canopy occlusion, and high fruit positions [12,24,25,26].

Each of the following three sections, Section 2.1, Section 2.2 and Section 2.3 follows a consistent format. First, the defining characteristics of the height layer are described. Next, the influence of growing conditions and plant morphology on manipulator design is examined. Then, current research is summarized, and the performance and limitations of key robotic prototypes are evaluated. By organizing the content by crop height, specific design needs for each layer are clarified, and both the technical differences between layers and the common challenges faced across them are highlighted. Below-soil crops (Section 2.4) are only included for comparative reference and boundary definition, and do not belong to the three core height layers of this review.

2.1. Ground-Level Crops (<0.6 m)

Strawberries are a representative example of ground-level crops, which hold significant economic value in greenhouse cultivation due to their high market demand. However, the automated harvesting of such low-stature crops has long been considered technically challenging because it requires a delicate balance among operational efficiency, fruit integrity, and adaptability to complex growing environments [17,18]. These difficulties stem from several common traits: compact planting layouts, narrow operational aisles, short and fragile stems, and soft fruits with thin peels that are prone to bruising. As a result, harvesting manipulators for ground-level crops must feature a small working radius, high-precision positioning, and damage-resistant end-effectors, while their overall size and mobility must be optimized to suit confined greenhouse spaces.

Occlusion handling and active removal. A major obstacle in strawberry harvesting is the dense foliage that often occludes the target fruit. To address this, several researchers have integrated active occlusion-removal mechanisms into their manipulators. In the following comparisons, “success rate” is broadly defined as the ratio of successfully harvested fruits to total attempts, while “harvesting success rate” refers to the harvesting success rate of the specific operation (e.g., the proportion of successfully picked fruits relative to total targets). As shown in Figure 4, He et al. [17] equipped a fan-integrated end-effector that blows away obstructive leaves after the vision system classifies strawberries into occluded and non-occluded categories, raising the harvesting success rate from 58.1% to 73.9%. Similarly, in the dual-arm system developed by Liu et al. [27], a brushless ducted fan is mounted on one arm to clear leaves while the other arm performs picking. More comprehensively, Parsa et al.