A New Underactuated Citrus-Picking End-Effector Achieves 98.3% Success Rate Across Fruit Shapes and Sizes
In the orchards of southern China, harvest season has long been synonymous with stooped backs, blistered hands, and a growing labor crisis that threatens the economic viability of citrus farming—not only in China but across the globe. As rural populations age and younger generations migrate to urban centers, the agricultural sector faces a stark reality: mechanical assistance is no longer optional. Enter a new generation of robotic harvesters, and at the forefront stands a breakthrough that may finally bridge the gap between laboratory promise and field-ready performance—a novel underactuated citrus-picking end-effector developed by researchers at the School of Advanced Manufacturing Engineering, Chongqing University of Posts and Telecommunications.
Unlike previous robotic grippers that often damaged fruit or failed on irregularly shaped specimens, this device marries mechanical elegance with intelligent control to adapt in real time to variations in size (30–100 mm in diameter) and ellipticity. In controlled lab trials simulating real-world conditions, it achieved a staggering 98.3% average picking success rate across three batches of citrus—spanning different cultivars, maturities, and degrees of deformation—with an average operation time of just 5.3 seconds per fruit and a damage rate of only 1.7%.
What makes this development noteworthy is not its raw power, but its passive and active compliance—a dual-layered strategy that mimics the nuanced dexterity of the human hand. At its core lies a dual-link parallel finger structure, inspired by the biomechanics of human grasping. Each of the three fingers integrates two sets of four-bar linkages: one actively driven by a motor, the other passively guided by a torsion spring and mechanical stop. This underactuated design (fewer motors than degrees of freedom) allows the fingers to self-conform to the fruit’s surface without complex sensing or computation. When a large orange is intercepted, all three phalanges sequentially wrap around it in a “enveloping grasp”; when a small, nearly spherical kumquat enters the workspace, the distal phalanx alone engages in a precise “pinch grasp”—a switch achieved purely through mechanical intelligence.
This passive compliance, however, is only half the story. Citrus fruits are rarely perfect spheres—many exhibit pronounced ellipticity, especially near the peduncle. When gripped symmetrically by rigid or non-rotating fingers, such irregular shapes generate torque imbalances: contact forces misalign, stress concentrates on the peel, and the fruit either slips or suffers micro-tears. The Chongqing team addressed this elegantly: they added a rotary joint at the base of each finger, allowing the entire finger assembly to swivel axially.
Crucially, this rotation is not pre-programmed. Instead, it is governed by an active compliance strategy based on real-time current feedback. As the fingers close and make contact, any mismatch between the finger surface and the local fruit curvature creates a lateral force component, which—through moment arms and geometry—translates into an axial torque at the finger root. The motor driving the rotary joint senses this as a deviation in current from a calibrated baseline. A PID controller then commands a corrective rotation until the torque (and thus the current offset) vanishes—indicating that the finger face is now locally normal to the fruit’s surface.
The result? Uniform contact, maximized frictional grip, and—most critically—no edge-induced scratching. This dual-mode compliance system—passive for size, active for shape—represents a significant departure from earlier solutions, which tended to over-engineer one aspect at the expense of the other. Soft-material grippers, for instance, mitigate surface damage but suffer from poor positioning repeatability and instability in high-wind orchard conditions. Rigid multi-finger hands with force sensors achieve high fidelity but require dozens of embedded transducers, complex calibration, and substantial onboard computation—cost and reliability barriers for commercial deployment.
The Chongqing end-effector sidesteps these pitfalls. Its kinematic chain is simple (just four motors: three for finger actuation, one for base rotation), its control logic is minimal (current deviation → rotation correction), and its structure is almost entirely 3D-printed in high-strength polymer, with soft silicone liners on the contact surfaces for added protection. The entire unit weighs under 850 grams—light enough to be mounted on standard collaborative robot arms without compromising payload or cycle time.
The team validated their design through both simulation and physical testing. Multibody dynamic simulations confirmed that the finger trajectories satisfied the kinematic constraints across the full range of target fruit sizes. Static force analysis, using the principle of virtual work, established a linear relationship between the motor’s output torque and the normal gripping force applied to the fruit—enabling precise control of grasp strength to avoid bruising. Most importantly, the hardware prototype was subjected to rigorous field-like trials: 348 citrus fruits were selected from local markets to represent the full spectrum of natural variation—some perfectly round, others oblong or slightly flattened, some with surface blemishes, others with stem remnants creating asymmetric protrusions.
In every case, the end-effector performed autonomously: it approached the fruit, initiated grasp, allowed the fingers to self-adapt, activated the current-feedback rotation, detected stability (via steady-state current), and then engaged the base motor to apply a gentle twisting motion—mimicking the “pull-and-twist” detachment employed by human pickers. The average success rate of 98.3% is not just a number; it reflects a system robust enough to handle outliers. Failures were almost exclusively attributed to pre-existing fruit decay (soft spots that yielded unpredictably under load), not design flaws. In no trial did the device scratch or puncture a sound fruit.
This robustness stems directly from the team’s user-centered design philosophy. Lead researcher Wei Bo notes in the paper’s introduction that “mechanization should enhance, not replace, the subtle skill of the human picker”—a principle embedded in every engineering decision. The enveloping-to-pinch transition mirrors natural hand motion. The torque-based finger alignment replicates the micro-adjustments a skilled worker makes without conscious thought. Even the final detachment method—rotational, not shear or pull—avoids damaging the vulnerable abscission layer, preserving post-harvest quality.
From an economic standpoint, the implications are profound. Citrus is among the most labor-intensive fruit crops; in some regions, harvesting accounts for over 47% of total production costs. A reliable, low-damage robotic picker not only cuts labor expenses but also reduces post-harvest losses due to handling injury—a major factor in fruit shelf life. Moreover, by standardizing detachment force and technique, the device could improve consistency in fruit quality, aiding growers in meeting strict export standards.
The work also opens new pathways for horticultural robotics beyond citrus. The underactuated finger design is highly scalable: by adjusting link lengths and spring constants, the same architecture can be adapted for apples, peaches, or even delicate produce like tomatoes and strawberries. The current-feedback rotation strategy, meanwhile, is universally applicable to any crop where surface geometry matters—eggplants, pears, or peppers, for instance.
Still, challenges remain before widespread field deployment. The current system relies on a fixed-base robotic arm; integrating it onto a mobile platform capable of navigating uneven orchard terrain is the logical next step. Vision systems for fruit detection and localization—while outside the scope of this paper—must be co-developed to enable true autonomy. Battery life, dust resistance, and rapid maintenance in muddy conditions are all practical considerations that future iterations must address.
Yet these are refinements, not rethinks. What the Chongqing team has delivered is a foundational advance—a mechanical and control architecture that is simultaneously simple, intelligent, and commercially viable. In a field often criticized for over-complication and under-delivery, their end-effector stands as a rare example of engineering humility: it doesn’t try to out-think nature; it learns from it.
As global food systems face mounting pressure from climate change, labor shortages, and rising consumer expectations for quality and sustainability, innovations like this will be critical. They represent not just a technical achievement, but a philosophical shift—from brute-force automation to collaborative augmentation, where machines extend human capability rather than erase it.
The citrus picker may be small, but its success suggests a larger truth: sometimes, the most powerful solutions are not those with the most sensors or the fastest processors, but those that listen—quite literally, in this case—to the subtle feedback of the world they seek to serve.
Wei Bo, He Jinyin, Shi Yang, Jiang Guangli, Zhang Xianyu, Ma Ying
School of Advanced Manufacturing Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
Transactions of the Chinese Society for Agricultural Machinery, 2021, 52(10): 120–128
DOI: 10.6041/j.issn.1000-1298.2021.10.012