Force Sensor-Based Picking Pattern Analysis and Two-Jaw Gripper Design for Cherry Tomato Robotic Harvesting in Greenhouse

Rajmeet Singh; Manveen Kaur; Yusuf Adamu; Irfan Hussain

cs.ROOct 13, 2026

Force Sensor-Based Picking Pattern Analysis and Two-Jaw Gripper Design for Cherry Tomato Robotic Harvesting in Greenhouse

Authors: Rajmeet Singh, Manveen Kaur, Yusuf Adamu, Irfan Hussain

Organizations: Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE · Khalifa University Center for Autonomous Robotic Systems (KUCARS), Khalifa University, Abu Dhabi, UAE · Department of Mechanical Engineering, University of Windsor, ON, Canada

Abstract

Robotic harvesting of cherry tomatoes in greenhouses remains challenging due to dense fruit clusters, delicate skin, and the need for gentle yet effective detachment. This study presents a force-data-driven framework for autonomous cherry tomato harvesting that integrates six-axis force/torque sensing, a bio-inspired compliant gripper, and deployment on a mobile robotic platform. A wearable ATI six-axis force/torque sensor mounted on the index finger captured kinetic signatures of three elemental picking patterns (gripping, twisting, and lateral pulling) across 150 manual trials (50 per pattern) in laboratory and greenhouse settings. Gripping required a peak normal force of 3.92 and 0.45 N, twisting produced a dominant torque of 19.20 2.10 Nmm about the pedicel axis with minimal truss disturbance, and pure lateral pulling alone was insufficient for reliable detachment. These measurements informed a novel fingertip-localised Finray gripper, with compliant TPU 95A Finray geometry restricted to the fruit-contact region and a rigid PLA body, emulating the dual-stiffness human fingertip. A Dynamixel XM430-W350 servomotor under current-based torque control applies the measured gripping force and a 45° wrist twist. Finite element analysis showed a maximum von Mises stress of 0.24 MPa under the 3.92 N load, below the bruising threshold of cherry tomato skin. Integrated on a UR5 arm and Clearpath Husky A200 base, the system was validated in NVIDIA Isaac Sim and in 50 autonomous trials at the SILAL commercial greenhouse, Abu Dhabi, UAE. The hybrid grip-twist-pull strategy achieved an 86.0% picking success rate, 2.3% fruit damage, and an 8.4 s cycle time per fruit, outperforming vacuum (66.3%), rotating (70.1%), clamping (76.0%), and suction-cutting (52.4%) end-effectors. These results demonstrate preliminary feasibility under the tested facility and cultivar conditions

Figures & tables

Figure 1: Typical planting pattern of cherry tomatoes in the SILAL commercial greenhouse, Abu Dhabi, UAE. (a) Overview of high-wire trellising system. (b) Close-up of cherry tomato truss with ripe fruits.

Category	Units	Mean	SD
Fruit diameter	(mm)	30.25	1.30
Fruit height	(mm)	26.10	1.12
Truss weight	(g)	168.06	15.50
Pedicel length	(mm)	43.52	6.44
Fruits number	(numbers)	15	3

Table 1: Means and standard deviations (SD) of physical parameters of cherry tomato fruit samples measured at SILAL greenhouse.

Figure 2: Cherry tomato picking patterns derived from manual harvesting observation at the SILAL greenhouse facility. (a) Pattern A: Grasping centripetal grip with axial pull ( $F_{pull}$ ) along the pedicel direction, causing tensile pedicel failure. (b) Pattern B: Twisting grip with progressive rotation ( $T_{z}$ ) about the pedicel axis, causing torsional shear failure. (c) Pattern C: Pulling direct horizontal pull ( $F_{pull}$ ) perpendicular to the truss axis, causing tensile and bending failure. (d) Hybrid (Proposed) the proposed robotic picking strategy combining progressive rotation ( $T_{z}$ ) with light lateral pull ( $F_{pull}$ ), applying normal forces ( $F_{n}$ ) axially to secure the fruit while minimising detachment force and truss disturbance. Blue arrows denote gripping forces ( $F_{g}$ ); orange arrows denote detachment forces and torques. Dashed ellipses indicate the abscission zone at the pedicel calyx junction.

Figure 3: Measurement setup of the ATI force sensor-based data acquisition system. The wearable ATI force sensor mounted on the finger is connected to the ATI Net Box controller, which receives a 24 V DC power supply and transmits the acquired force and torque data to a personal computer via a LAN (Ethernet) connection for real-time monitoring, processing, and storage.

Physical Properties
Parameter	Range	Resolution	Overload Limit
$F_{x}$ , $F_{y}$	$\pm$ 145 N	1/16 N	$\pm$ 300 N
$F_{z}$	$\pm$ 290 N	1/8 N	$\pm$ 900 N
$T_{x}$ , $T_{y}$	$\pm$ 5 N $\cdot$ m	1/376 N $\cdot$ m	$\pm$ 9 N $\cdot$ m
$T_{z}$	$\pm$ 5 N $\cdot$ m	1/376 N $\cdot$ m	$\pm$ 9 N $\cdot$ m
Diameter	43 mm

Table 2: ATI force/torque sensor specifications.

Figure 4: ATI force sensor finger-mounted assembly and force diagram during tomato picking. (a) Front view of the 3D-printed finger mounting bracket secured with adjustable straps for stable attachment to the finger during picking trials. (b) Side view illustrating the complete assembly comprising the 3D-printed finger mount, ATI force sensor (FT43304), and silicon tip, with the three-axis coordinate frame ( $x$ , $y$ , $z$ ) indicating the sensing directions and a 2 mm reference scale shown for dimensional context. (c) Isometric view of the ATI force sensor integrated within the 3D-printed mount, with the coordinate frame axes ( $x$ , $y$ , $z$ ) denoting the force and torque measurement directions. (d) Free-body diagram of the applied forces during a two-finger tomato picking trial, showing the normal forces ( $F_{\mathrm{n,f}}$ and $F_{\mathrm{n,t}}$ ) and frictional forces ( $F_{\mathrm{f,f}}$ and $F_{\mathrm{f,t}}$ ) exerted by the index finger and thumb on the tomato, respectively, along with the gravitational force ( $mg$ ) acting at the centre of gravity (COG) of the tomato.

Figure 5: Experimental setup for cherry tomato picking force measurement. (a) Three proposed hybrid picking motion patterns applied during the harvesting trials: normal force to grip (lateral pinch gripping), twisting (rotational detachment), and axial pull (straight stem-pull detachment), each performed with the ATI force sensor mounted on the index finger to capture the corresponding six-axis force and torque profiles. (b) Sample images of picking experimentation conducted in the greenhouse environment, demonstrating the wearable ATI force sensor deployment during actual cherry tomato harvesting on the vine under real growing conditions. (c) Controlled laboratory experimentation using a custom 3D-printed test bench, comprising a rigid U-shaped 3D-printed structure with screw-based clamping to fix the cherry tomato truss in a repeatable position, enabling systematic force measurement with the wearable ATI force sensor under standardised conditions.

Figure 6: Proposed novel fingertip-localised Finray gripper for damage-free cherry tomato harvesting. (a) CAD model of the gripper assembly illustrating the major structural components, including the outer protective cover, internal support frame, and the Finray-inspired compliant fingertips fabricated from flexible TPU 95A material for gentle conformal contact with the tomato surface. (b) Fabricated prototype of the gripper, with the rigid structural components 3D-printed in PLA material and actuated by a Dynamixel XM430-W350 smart servomotor through an integrated Dynamixel controller mounted on the top surface, enabling closed-loop control of the gripping force and jaw opening/closing motion.

Parameter	Value
Stall Torque (12 V)	4.1 N $\cdot$ m
No-Load Speed	46 RPM
Operating Voltage	10.0–14.8 V
Resolution	4096 pos/rev (0.088 ∘ )
Communication	TTL / RS-485 (Protocol 2.0)
Weight	82 g

Table 3: Dynamixel XM430-W350 servo motor specifications.

Figure 7: Bio-inspired design rationale and finite element method (FEM) validation of the proposed fingertip-localised Finray gripper for damage-free cherry tomato harvesting. (a) Conceptual analogy between the human fingertip and the proposed gripper, illustrating the biomimetic dual-stiffness configuration in which the soft, deformable distal fingertip pad is replicated by the compliant TPU 95A Finray tip (highlighted in red circle), while the rigid supporting phalangeal structure is emulated by the PLA-printed rigid body of the gripper (highlighted in red rectangle). (b) Von Mises stress distribution under an applied gripping force of 3.92 N derived from the average peak force measured across 50 manual cherry tomato picking trials using the wearable ATI force sensor on the index finger showing a maximum stress of 0.24 MPa localised within the Finray rib structure, well below the yield strength of TPU 95A. (c) Corresponding total displacement distribution showing a maximum tip deflection of 1.566 mm at the contact region, enabling passive conformal adaptation to the tomato curvature.

Figure 8: Complete mobile harvesting platform. Husky A200 base with UR5 manipulator and gripper deployed at greenhouse.

Component	Model	Key Specification
Force/Torque Sensor	ATI	6-axis, $\pm$ 145/290 N
Gripper Motor ( $\times$ 2)	Dynamixel XM430-W350	4.1 N $\cdot$ m, 82 g
Manipulator	Universal Robots UR5	850 mm reach, 5 kg payload
Mobile Base	Clearpath Husky A200	75 kg payload, 1.0 m/s
Close-Range Camera	Intel RealSense D405	Depth 7–100 cm
Wide-Angle Camera	Intel RealSense D455	Depth 40 cm–6 m

Table 4: Complete hardware specifications for the mobile cherry tomato harvesting platform.

Figure 9: Validation of the proposed autonomous cherry tomato harvesting pipeline in the NVIDIA Isaac Sim environment integrated with ROS2. (a) Simulation setup showing the UR5 manipulator equipped with the proposed gripper and end-effector-mounted RealSense depth camera, deployed within a photorealistic greenhouse scene populated with tomato plant models and a designated fruit collection bin. (b) Real-time ripe tomato detection and picking-point localisation performed using the proposed YOLO26-RipeLoc Lite model, where each detected ripe fruit is annotated with its ripeness confidence score (e.g., RIPE-3, RIPE-4, RIPE-5) and a centre point marker indicating the target picking location for the manipulator. (c) Sequential simulation snapshots illustrating the complete autonomous harvesting pipeline in Isaac Sim, including target approach, gripper alignment, hybrid grip–twist–pull execution, fruit detachment, and placement in the collection bin.

Figure 10: Six-axis force and torque time-series profiles measured during manual cherry tomato picking trials using the wearable ATI force sensor mounted on the index finger. (a, d) Normal force to grip (clamping) pattern showing sustained compressive $F_{z}$ of $-3.92$ N and bending torque $T_{x}$ peaking at $8.66$ N $\cdot$ mm during the steady-state grip phase ( $t=1.3$ – $4.0$ s). (b, e) Twisting pattern exhibiting the highest compressive $F_{z}$ of $-5.09$ N coupled with a dominant rotational torque $T_{y}$ of $19.20$ N $\cdot$ mm about the pedicel axis. (c, f) Axial pull pattern showing smooth progressive force build-up with peak $F_{z}$ of $0.22$ N and transient torque peaks during the initial engagement phase. The distinct kinetic signatures of the three picking patterns provide the empirical basis for designing the proposed hybrid grip–twist–pull autonomous harvesting algorithm.

Pattern	Peak $F_{z}$ (N)	Peak $T_{y}$ (N $\cdot$ mm)	Duration (s)
A – Gripping	$3.92\pm 0.45$	$6.48\pm 0.70$	$2.6\pm 0.3$
B – Twisting	$5.09\pm 0.58$	$19.20\pm 2.10$	$4.2\pm 0.5$
C – Axial Pull	$0.22\pm 0.03$	$0.71\pm 0.09$	$3.1\pm 0.4$

Table 5: Comprehensive comparison of the three picking patterns across evaluation criteria, derived from ATI Mini45 force/torque measurements during manual cherry tomato picking trials. Values are reported as mean

\pm

standard deviation across the trial duration.

Figure 11: Temporal evolution of UR5 joint angles ( $J_{0}$ – $J_{5}$ ) during one complete autonomous cherry tomato harvesting cycle executed by the proposed hybrid grip–twist–pull algorithm in the Isaac Sim simulation environment. The seven sequential phases of the cycle are colour-shaded: APPROACH (grey), GRIP (yellow), TWIST (red), PULL (blue), TRANSPORT (green), RELEASE (purple), and HOME (light grey). The characteristic $+45^{\circ}$ wrist roll rotation executed by joint $J_{5}$ during the TWIST phase (brown trace, $180^{\circ}\rightarrow 225^{\circ}$ ) is clearly resolved, demonstrating the precise execution of the designed twist primitive while all other joints remain stationary. The largest joint excursion occurs at the base joint $J_{0}$ during the TRANSPORT phase, where the fruit is delivered to the collection bin.

End Effector	Tests	Success (%)	Damage (%)	Time (s)
Vacuum [ Gao et al., 2024a ]	95	66.3	0.0	5.3
Rotating [ Gao et al., 2024a ]	97	70.1	5.2	6.4
Clamping [ Xie et al., 2022 ]	75	76.0	High	N/A
Suction-cutting [ Fujinaga et al., 2021 ]	105	52.4	9.5	23.0
Ours — Hybrid Grip–Twist–Pull	50	85.0	2.0	8.4

Table 6: Picking performance comparison between the proposed fingertip-localised Finray gripper with hybrid grip–twist–pull strategy and benchmark end-effectors reported in the literature for cherry tomato harvesting.

Figure 12: Sequential snapshots of one complete autonomous cherry tomato harvesting cycle conducted in the SILAL commercial greenhouse facility (Abu Dhabi, UAE) using the proposed UR5 manipulator equipped with the fingertip-localised Finray gripper. The upper row shows the APPROACH phase where the YOLO26-RipeLoc Lite detector identifies the target ripe fruit (yellow bounding box) and the UR5 navigates through the plant canopy culminating in the GRIP phase, where the Finray jaws close on the target fruit at the calibrated force of $F_{grip}=3.92$ N. The lower row illustrates the TWIST phase ( $+45^{\circ}$ wrist roll about the pedicel axis), the PULL phase (low-velocity axial separation), the TRANSPORT phase (delivery of the harvested fruit to the collection bin), and the final RELEASE phase. The sequential frames demonstrate the smooth and damage-free execution of the proposed hybrid grip–twist–pull algorithm under realistic, unstructured greenhouse operating conditions.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Symbol	Description	Unit
$F_{x},F_{y},F_{z}$	Force components along $x$ , $y$ , $z$ axes	N
$T_{x},T_{y},T_{z}$	Torque components about $x$ , $y$ , $z$ axes	N $\cdot$ mm
$F_{\mathrm{n,f}}$	Normal force by index finger	N
$F_{\mathrm{n,t}}$	Normal force by thumb	N
$F_{\mathrm{f,f}}$	Frictional force at finger contact	N
$F_{\mathrm{f,t}}$	Frictional force at thumb contact	N

Appendix

Table 7: Nomenclature: Force and torque variables.

Abbrev.	Full Form
ATI	ATI Industrial Automation
CAD	Computer-Aided Design
COG	Centre of Gravity
DOF	Degree of Freedom
FDM	Fused Deposition Modelling
FEM	Finite Element Method

Appendix

Table 8: Nomenclature: Abbreviations.

Robotic Harvesting

See all 22 papers

Feb 22, 2026

FruitTouch: A Perceptive Gripper for Gentle and Scalable Fruit Harvesting

The automation of fruit harvesting has gained increasing significance in response to rising labor shortages. A sensorized gripper is a key component of this process, which must be compact enough for confined spaces, able to stably grasp diverse fruits, and provide reliable feedback on fruit conditions for efficient harvesting. To address this need, we propose FruitTouch, a compact gripper that integrates high-resolution, vision-based tactile sensing through an optimized optical design. This configuration accommodates a wide range of fruit sizes while maintaining low cost and mechanical simplicity. Tactile images captured by an embedded camera provide rich information for real-time force estimation, slip detection, and softness prediction. We validate the gripper in real-world fruit harvesting experiments, demonstrating robust grasp stability and effective damage prevention.

Ruohan Zhang, Mohammad Amin Mirzaee, Wenzhen Yuan

University of Illinois at Urbana-Champaign, Champaign, IL, USA

Robotic Grasping Robotic Hand Design Agricultural Robotics Vision-Based Tactile Sensing

Ask PDF

Apr 27, 2026

An analysis of sensor selection for fruit picking with suction-based grippers

Robotic fruit harvesting often fails to reliably detect whether a fruit has been successfully picked, limiting efficiency and increasing crop damage. This problem is difficult due to compliant fruit and grippers, variable stem attachment, and occlusions in orchard environments. Prior work has explored vision-based perception and multi-sensor learning approaches for pick state estimation. However, minimal sensor sets and phase-dependent sensing strategies for accurate pick and slip detection remain largely unexplored. In this work, we design and evaluate a multimodal sensing suite integrated into a compliant suction-based apple gripper. Our approach is unique because it identifies which sensors are most informative at different phases of the pick, enabling predictive detection of failures before they occur. The contributions of this paper are a phase-dependent evaluation of multimodal sensors and the identification of minimal sensor sets for reliable pick state classification. Experiments in a real apple orchard show that Random Forest and Multilayer Perceptron classifiers detect successful picks and impending failures with over 90% accuracy, and Random Forest predicts pick/slip events within 0.09 s of human-annotated ground truth.

Eva Krueger, Marcus Rosette, Joseph R. Davidson

Collaborative Robotics and Intelligent Systems (CoRIS) Institute, Oregon State University, Corvallis OR 97331, USA

Robotic Grasping Tactile Sensing Agricultural Robotics Multimodal Sensor Fusion

Ask PDF

Sep 16, 2026

CLASP: A Cluster-Level Autonomous Selective Picking Robot with a Soft Rolling-Band Gripper for Fresh-Market Blueberry Harvesting

Fresh-market blueberries require selective, gentle picking, which is labor-intensive and expensive. Over-the-row machine harvesters are fast but non-selective, bruising mixed-ripeness fruit and limiting yield to the processing market. Selective robotic harvesters typically target individual fruits rather than fruit clusters, which limits harvesting efficiency for small, densely clustered blueberries. This paper presents CLASP, a Cluster-Level Autonomous Selective Picking robot with a Soft Active Rolling-Band Gripper (SARB-Gripper). Two compliant bands envelop the cluster and roll against the fruit, drawing mature berries off in sequence, while closed-loop regulation of the pulling force keeps the applied load below the immature detachment threshold. A global-to-local perception pipeline pairs an eye-to-hand camera for global cluster detection and target selection with an eye-in-hand camera for local localization and cluster orientation estimation. Field measurements confirm a clear detachment-force separation between mature and immature fruit, and the SARB-Gripper reproduces a commanded pulling force to within \SI{3.7}{\percent}, enabling selective harvesting at the cluster level. In end-to-end field trials, CLASP autonomously grasped 23 of 25 presented clusters (\SI{92}{\percent}). With the component cost of approximately $3326 per unit, CLASP offers a scalable approach to selective cluster-level harvesting for fresh-market blueberries.

Yixuan Xia, Yilin Cai, Natalia Belen Espinoza +4

George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332 USA · Department of Crop & Soil Sciences, University of Georgia, Athens, GA 30605, USA · Bio-Sensing, Automation, and Intelligence Laboratory, Department of Agricultural and Biological Engineering, University of Florida, Gainesville, Florida 32611, USA +3

Robotic Grasping Force-Controlled Robotic Manipulation Agricultural Robotics Soft Robotics Robotic Harvesting

Ask PDF

Abstract

Robotic Harvesting

Explore similar work

FruitTouch: A Perceptive Gripper for Gentle and Scalable Fruit Harvesting

An analysis of sensor selection for fruit picking with suction-based grippers

CLASP: A Cluster-Level Autonomous Selective Picking Robot with a Soft Rolling-Band Gripper for Fresh-Market Blueberry Harvesting