Robotiq 2F-85 Gripper Overview

Updated 16 January 2026

The Robotiq 2F-85 Gripper is an adaptive, two-fingered parallel-jaw gripper designed for precision grasping in collaborative robotics.
Featuring a programmable 0–85 mm stroke and up to 100 N force, it integrates tactile sensors and customizable feedback for reliable performance.
It supports advanced grasping algorithms and manipulation policies, achieving high success rates in simulation and real-world industrial trials.

The Robotiq 2F-85 Gripper is an adaptive, two-fingered parallel-jaw industrial gripper extensively employed in collaborative robotics, robotic grasping research, large-scale dataset curation, and tactile sensing prototyping. Characterized by its programmable stroke, modular end-effector interface, and robust actuation, the 2F-85 is a staple in environments demanding repeatable precision, adaptive surface contact, and software-driven integration. Its application spectrum includes physics-based grasp evaluation, zero-shot cross-gripper manipulation, high-throughput grasp trial automation, and closed-loop teleoperation with force feedback.

1. Physical and Electrical Specifications

The Robotiq 2F-85 Gripper features an adjustable opening (stroke) of 0–85 mm, supporting object widths within this range. Its maximum programmable gripping force is 100 N (user-selectable from 5 to 100 N), delivered via two adaptive curved fingers with a repeatability of ±0.1 mm (Proesmans et al., 2023). The gripper weighs ~1.3 kg and is equipped with an ISO 9409-1-50-4-M6 flange for compatibility with major industrial robot wrists. Electrical supply is 24 V DC drawn from the robot’s tool output, necessitating a standby current of 150 mA and peaking at 3 A during finger movement. Standard I/O includes digital lines for basic open/close control and analog voltage feedback (0–10 V) for real-time position monitoring. Integrated safety features such as a resettable fuse (2 mA) and P-channel safety MOSFET ensure electrical isolation and fault tolerance.

2. Geometry and Kinematic Modeling

Gripper kinematics are dictated by a single actuated degree of freedom enabling symmetric closing of the fingers. Each finger is typically modeled as a rigid plate (22 mm pad thickness) in simulation and collision engines. In grasp generation pipelines, the 2F-85 is parameterized by a 6-DoF rigid body transform $T \in SE(3)$ for the palm frame and a scalar jaw width $d \in [0,85]$ mm, formally $g = (T, d)$ (Srinivas et al., 24 Sep 2025). The palm frame is defined such that its origin coincides with the midpoint between the fingertips when centered, with $z$ -axis as the approach direction and $y$ -axis bisecting the finger slot. No further finger articulation or Denavit–Hartenberg parameters are sampled in leading datasets or simulation environments; fingers remain parallel throughout closure (Srinivas et al., 24 Sep 2025).

On the Franka Panda flange, the tool-center-point (TCP) offset of the 2F-85 is $\Delta_{CTP} \approx 0.205$ m along the tool- $Z$ axis (Yao et al., 21 Feb 2025). The effective jaw-width constraint for manipulation policies is $w_{min}=7.5$ cm and $w_{max}=11.5$ cm. Control architectures leverage affine mappings from motor stroke to fingertip separation ( $d_f(s) = d_0 + \alpha s$ , with $d_0 \approx 0$ and $\alpha \approx 1.0$ ) (DuFrene et al., 2024).

3. Sensing and Integration: Tactile, Force, Wire-Free Architectures

A key feature for research is seamless integration of tactile sensors. Microcontroller-based sensor readouts, such as the “Halberd” coupling board, allow direct attachment of resistive, capacitive, or gel-based tactile modules via modular headers, matched to the Arduino Nano 33 BLE pinout (Proesmans et al., 2023). The architecture incorporates a u-blox Nina B301 microcontroller (ARM Cortex-M4F, BLE/Wi-Fi), regulated supply chains (24 V → 5 V → 3.3 V), and safety circuitry to support robust, plug-compatible sensor deployment. Up to 8 analog (12-bit) and 15 digital I/O pins are exposed, and all data can be streamed over BLE, Wi-Fi, or USB, with typical end-to-end latencies <10 ms.

For force feedback, customized 3D-printed fingers with embedded force-sensitive resistors (RP-C7.6-LT) are bolted directly into the jaw mount points. The output from these sensors is linearized via an op-amp current-to-voltage converter: $V_{OUT} = V_{REF} \times \left(-\frac{R_G}{R_{FS}}\right)$ , where $V_{REF} = 3.3$ V is the reference voltage, $R_G$ the feedback gain resistor, and $R_{FS}$ the sensor resistance at full compression (Satsevich et al., 1 Oct 2025). The resulting voltage is converted to force via a manufacturer-supplied curve, and real-time data is relayed via RS-485 to the robot controller, enabling haptic teleoperation and dataset enrichment.

4. Grasping Algorithms, Dataset Integration, and Physics Evaluation

The Robotiq 2F-85 is the subject of large-scale object-centric grasping datasets such as GraspFactory, which comprises 97.1 million validated grasp candidates (physics-evaluated) over 33,710 unique objects (Srinivas et al., 24 Sep 2025). Raw grasp candidates are generated using antipodal sampling, mesh decimation for contact diversity, and rigid-body collision filtering. Feasibility is defined by the absence of finger-object interpenetration and physics-sustained retention during simulated perturbations. Clustering procedures in $SE(3)$ space and explicit constraint enforcement during inference facilitate integration into machine learning models. Representative equations for grasp parameterization and clustering include:

Grasp transform: $T = \begin{bmatrix} R & t \ 0 & 1 \end{bmatrix}$ , $R \in SO(3)$ , $t \in \mathbb{R}^3$ .
Grasp tuple: $g = (T, d)$ , $T \in SE(3)$ , $d \in [0,85]$ mm.
Clustering distance: $d(g_1,g_2)=\|t_1-t_2\|+\arccos(|q_1\cdot q_2|)$ .

Empirical evaluation with the UR-10e + 2F-85 yielded real-world grasp success rates per part between 80% (Base, Pose 1) and 100% (Wheel, Hanger), with robustness on four parts averaging 96%-100% in repeated trials (Srinivas et al., 24 Sep 2025).

5. Manipulation Policy Transfer and Cross-Gripper Optimization

Diffusion-based policy frameworks, such as those presented in the hybrid learning-optimization paradigm, demonstrate zero-shot adaptation for pick-and-place tasks across multiple gripper types, including the 2F-85. At inference, gripper-specific constraints—TCP offset, jaw width range, signed-distance-field margins—are enforced via gradient-projection steps appended to the reverse diffusion process (Yao et al., 21 Feb 2025). The constraint update is

$x_{t-1}^* \leftarrow \hat{x}_{t-1} - \eta_{tcp} \nabla_x C_{tcp}(\hat{x}_{t-1}) - \eta_{jaw} \nabla_x C_{jaw}(\hat{x}_{t-1}) - \eta_{coll} \nabla_x C_{coll}(\hat{x}_{t-1}),$

guaranteeing kinematic compliance and collision safety. The manipulation primitives (pre-grasp, grasp closure, lift/transfer, release) are retargeted to the 2F-85 by TCP waypoint adjustment and jaw width rescaling. Empirical trials reveal 96.7% average task success rate with the 2F-85 versus 24.3% for unconstrained policies, with mean success in pick and place operations of 98.3% and 95.0% respectively (Yao et al., 21 Feb 2025).

6. Automated Grasping Trials and State-Machine Control

High-throughput grasp testing is facilitated by the Grasp Reset Mechanism (GRM), wherein the 2F-85 is directly mounted to robots such as the Kinova Gen3 via a custom adapter. State-machine control is implemented via ROS and FlexBE, enabling open/close, speed-force parameterization, and feedback logging. In GRM-executed grasps, adaptive compliance and programmable force limits are leveraged for both delicate and rigid objects (DuFrene et al., 2024). Performance analysis over 1,020 trials reports a 70% overall success rate, with side grasps generally more robust than top grasps to spatial and orientational perturbations. Motor current is monitored during closure to implement a binary force-limit feedback loop for grasp success/failure characterization.

7. Mechanical Modifications and Teleoperation Enhancements

Custom finger modules extend the capabilities of the stock Robotiq 2F-85, most notably for tactile sensing and force feedback. Soft-touch silicone pads, rod–spring–guide mechanisms, and integrated resistive sensors augment compliance and load uniformity (Satsevich et al., 1 Oct 2025). Such designs improve manipulation performance for deformable or slippery objects by reducing incident peak forces (35.77% reduction observed in teleoperated egg pick-and-place). Mechanical adaptation enables compatibility with real-time force-feedback teleoperation and large-scale vision-language-action policy dataset collection with success rates on deformables near 90%. The modularity of the finger integration permits straightforward adaptation to alternative robot arms or gripper variants.

All technical content, experimental metrics, equations, and methodology here directly reflect explicit details from published arXiv works (Proesmans et al., 2023, Srinivas et al., 24 Sep 2025, Yao et al., 21 Feb 2025, DuFrene et al., 2024, Satsevich et al., 1 Oct 2025). No claims or implementations are inferred beyond furnished data.