Efficient Force and Stiffness Prediction in Robotic Produce Handling with a Piezoresistive Pressure Sensor

Properly handling delicate produce with robotic manipulators is a major part of the future role of automation in agricultural harvesting and processing. Grasping with the correct amount of force is crucial in not only ensuring proper grip on the object, but also to avoid damaging or bruising the product. In this work, a flexible pressure sensor that is both low cost and easy to fabricate is integrated with robotic grippers for working with produce of varying shapes, sizes, and stiffnesses. The sensor is successfully integrated with both a rigid robotic gripper, as well as a pneumatically actuated soft finger. Furthermore, an algorithm is proposed for accelerated estimation of the steady-state value of the sensor output based on the transient response data, to enable real-time applications. The sensor is shown to be effective in incorporating feedback to correctly grasp objects of unknown sizes and stiffnesses. At the same time, the sensor provides estimates for these values which can be utilized for identification of qualities such as ripeness levels and bruising. It is also shown to be able to provide force feedback for objects of variable stiffnesses. This enables future use not only for produce identification, but also for tasks such as quality control and selective distribution based on ripeness levels.

💡 Research Summary

The paper presents a low‑cost, quickly fabricated piezoresistive pressure sensor and demonstrates its integration with both rigid and pneumatically actuated soft robotic grippers for handling a variety of produce. The sensor is built from Velostat, a carbon‑infused polymer, patterned with copper‑tape electrodes using a consumer‑grade vinyl cutter. Two designs (a 3×2 array of larger pixels and a 2×2 array of smaller pixels) can be produced in under 15 minutes for less than one US dollar in raw materials, and the pixel geometry can be customized to fit different end‑effector shapes. Electrical readout is performed with an Arduino Mega 2560, two multiplexers, and a 10‑bit ADC, yielding a nominal sampling rate of about 15 Hz (upgradable by increasing the serial baud rate).

When pressure is applied, Velostat’s resistance decreases; the sensor therefore reports a relative resistance change with respect to a baseline measured in the absence of load. Absolute resistance values drift with temperature and humidity, but normalizing each pixel’s response eliminates most of this environmental dependence. Long‑term cycling (2 500 cycles over 13 h) reveals an initial “break‑in” period where sensitivity changes, after which the response stabilizes.

A key contribution is the algorithm for rapid estimation of the settled resistance value. The raw transient exhibits a sharp spike followed by an exponential decay. The authors fit a three‑parameter model R(t)=A·e^(−Bt)+C using three points: the cutoff time (t_c), the peak resistance time (t_p) within that window, and a small offset (Δt = 0.5 s) after the peak to avoid high‑frequency noise. By varying the cutoff time from 0.5 s to 20 s and comparing the fitted steady‑state value to the true settled resistance (measured after 20 s), they find that a 2.5 s cutoff yields a median error of ≈2.5 %, far better than the ≈6 % error obtained by simply using the raw resistance at the same time. This reduces the waiting period for a reliable measurement by a factor of four while preserving sufficient accuracy for force‑feedback control.

The sensor is mounted on a LocoBot WX250S platform equipped with a pair of servo‑driven rigid fingers, as well as on a pneumatically actuated soft finger. In both configurations the average normalized resistance change is mapped to grasping force using a calibration curve derived from a load cell. Simultaneously, the pressure distribution across the pixel array provides an estimate of the object’s deformation, which is converted into an effective stiffness (elastic modulus) value. Experiments with silicone sheets of four different stiffnesses and with real produce (apples, tomatoes, strawberries) demonstrate that the system can resolve force differences on the order of 0.1 N and stiffness variations of ≈0.05 MPa.

Beyond force control, the authors explore three application scenarios: (1) size estimation by counting activated pixels and the magnitude of resistance drop, (2) ripeness assessment by correlating measured stiffness with known firmness curves for specific fruits, and (3) bruise detection by identifying abnormal transient spikes or deviations from expected steady‑state values. These capabilities suggest that a single sensor can support both manipulation and quality‑assessment tasks in agricultural robotics.

The paper’s contributions are: (i) a truly low‑cost, easily reproducible piezoresistive pressure sensor, (ii) an exponential‑decay fitting method that provides near‑steady‑state readings within a few seconds, (iii) successful integration with both hard and soft grippers, and (iv) demonstration of force, stiffness, size, ripeness, and damage inference from the same data stream. Limitations include the modest 15 Hz sampling rate, pixel‑to‑pixel resistance variability that requires per‑pixel calibration, and the lack of an adaptive environmental compensation scheme for long‑term deployments. Future work is proposed to incorporate higher‑speed ADCs, wireless data links, multi‑sensor arrays for three‑dimensional pressure mapping, and machine‑learning models that fuse force‑stiffness data with visual cues for robust produce classification.