Intelligent Mechanical Characterization of Date Fruits for Automated Harvesting Grippers

Robotic harvesting of date fruits requires precise grasping force control to prevent tissue damage, yet cultivar- specific biomechanical limits remain absent from the literature. This work presents the first continuous stress-strain charac- terization of three Saudi date cultivars (Ajwa, Barhi, Sagai) across three hydration states, translated into validated robotic grasping constraints. A custom parallel-plate compression sys- tem emulates two-finger robotic grasping, while a Mask R-CNN vision model provides non-contact geometric measurement with below 5% relative error. Cyclic loading experiments establish elastic strain limits, with conservative operational thresholds of 7% for Ajwa and Barhi and 5% for Sagai. Validation with a UR10e manipulator confirms damage-free manipulation across all cultivars residual deformation below 1 mm and strain tracking error below 1%. Future work will integrate vision and force feedback with machine learning models trained on these experimentally derived limits, enabling real-time geometry- based gripper control for fully autonomous harvesting. Introduction and Related Work Saudi Arabia leads global date exports ($390M USD annually, 30M palm trees) [1], yet manual harvesting exposes workers to serious hazards including falls, with an annual fa- tality rate of 3.4 per 1,000 workers [2]. Automated harvesting is therefore both an economic and safety imperative [3]. Vision-based ripeness detection [4], [5] and cluster-level robotic harvesters [6], [7] have been demonstrated, but none address safe manipulation of individual separated fruits. AI- driven systems for apples, and kiwi [8], [9] show that data-driven gripping can reduce damage when physically grounded force limits are available. Force-aware gripper designs soft actuators [10], [11], force controllers [12], [13], and contact-pressure analyses [14]– [16], all depend on cultivar-specific mechanical limits that do not yet exist for date fruits. Prior date-fruit studies char- acterize physical dimensions [17], [18] and report discrete hardness or modulus values [19]–[22], but continuous stress- strain curves, cyclic elastic-recovery verification, and bioyield thresholds are absent from the literature. This work fills that gap. Methodology A custom parallel-plate uniaxial horizontal apparatus (50 N capacity, ±0.5% tolerance, 100 Hz [23]) emulates two- finger grasping and resolves the low-strain elastic regime that standard UTMs cannot reliably capture in compliant tissue. A Mask R-CNN (ResNet-50 FPN) model fine-tuned on a custom date dataset provides frame-by-frame segmentation during compression. Active-contour post-processing refines each mask to the true boundary; validation yields MAE of 0.991 cm2 in area (4.58%, R2>0.96) and 2.17 mm in diameter (2.19%), giving combined stress uncertainty ≈5%. 450 samples (50 per cultivar × hydration-state cell) at the Tamar ripeness stage were sized by SRS at 95% confidence and 2.67% relative margin of error ( ̄y=8.99 N, s=2.59 N). Moistened fruits were immersed at 40◦C for 3-4 h; dried fruits were heated to 80◦C for 8-12 h [24]. Sequential load-unload cycles at 1, 3, 5, . . . , 17% strain were applied to three fruits per cultivar. Residual deformation δr<0.1 cm (below visual detection, ≈3-4% of mean diame- ter) was classified as elastic; the bioyield point was the first strain level at which δr exceeded this threshold consistently. Linear regression on gripper-gap vs. command data yielded Gmax=77.59 mm, k=0.883 mm/%. The robot command to impose strain ε on a fruit of initial diameter L0 is per = l 77.59−L0(1−ε) 0.883 m . Results and Discussion Two-way ANOVA showed cultivar and hydration state significantly affected all dimensions (p<0.001). Barhi had the highest moisture sensitivity (54.9% volumetric reduction, normal to dried); Ajwa was most stable (9.6% diameter change); Sagai was intermediate [25]. Dried samples were stiffest in all cultivars. Ajwa dried specimens reached 20.50±2.50 kPa at 39% strain versus 12.20±0.74 kPa moistened (68% difference). Ajwa showed a stress plateau at 15-23% strain, supporting predictable-force grasping. Barhi had the highest variability (SE 10.3% of mean at 23% strain). Cyclic tests gave bioyield points of ≈9% for Ajwa (CV 6.96%) and Barhi (CV 32.96%), and 7% for Sagai (CV 11.90%). Barhi’s high variability necessitates a conser- vative 7% operational limit (Table I). TABLE I: Mechanical properties at the bioyield point (mean ± SE). Property Ajwa Barhi Sagai Force (N) 1.70±0.11 1.19±0.09 0.82±0.05 Stress (kPa) 3.29±0.20 2.16±0.15 1.72±0.11 Safe strain 7% 7% 5% Young’s mod. (kPa) 47.0±2.79 30.8±2.20 34.5±2.26 A supervised experimental validation using robotic ma- nipulator completed pick-lift-release cycles for all cultivars with no bruising, puncture, or flesh compression (Fig. 1). ICRA2026 Late Breaking Results Poster presented at 2026 IEEE International Conference on Robotics and Automation (ICRA 2026) June 1-5, 2026. Vienna, Austria