Robot End Effector Selection Guide: Types, Specs & Compatibility

What Is a Robot End Effector?

A robot end effector is any device attached to the end of a robot arm that interacts with the environment. It is the point where the robot meets the physical world. While the arm provides reach, positioning, and force, the end effector determines what the robot can actually do — grasp objects, apply adhesive, weld joints, insert screws, or sense surfaces.

End effectors matter more than most teams realize. A well-chosen end effector can make a simple 6-DOF arm solve tasks that a poorly equipped 7-DOF arm cannot. Conversely, the wrong end effector is the most common reason a robot manipulation project fails in practice. A robot arm is only as capable as the tool at its tip.

The term "end effector" encompasses all tool types: grippers (parallel jaw, three-finger, soft), suction cups, magnetic pickups, welding torches, spray guns, deburring tools, inspection cameras, force-torque sensors, and dexterous hands. This guide focuses on manipulation end effectors — tools designed to grasp, hold, and place objects — because they represent over 80% of end effector applications in robotics research and light manufacturing.

The 7 Main End Effector Types

Deep Dive: Parallel Jaw Grippers

The parallel jaw gripper is the workhorse of industrial and research robotics. Two flat or contoured fingers move linearly toward each other, clamping an object between them. The mechanism is simple — a single actuator (electric motor, pneumatic cylinder, or servo) drives both jaws symmetrically.

When to use: Any task involving picking rigid objects with at least two parallel surfaces for the jaws to grip. This covers the vast majority of manufacturing parts, packaged goods, and laboratory items. If your object can be picked up with two fingers of a human hand, a parallel jaw gripper will likely work.

Key specifications to evaluate:

• Stroke (opening width): The maximum distance between jaws. The Robotiq 2F-85 has an 85 mm stroke; the 2F-140 has 140 mm. Choose a stroke that accommodates your largest object plus 10-15 mm clearance.
• Grip force: The maximum clamping force. Ranges from 5 N (delicate grippers) to 200+ N (industrial). Match to your heaviest object: required grip force = (object weight x safety factor x 2) / friction coefficient. For rubber-coated fingers on smooth metal, friction ~0.6, safety factor 2x.
• Finger speed: How fast the gripper opens and closes. Industrial grippers close in 50-200 ms. For high-throughput pick-and-place, this matters — a 200 ms close time at 10 picks/minute costs 33 seconds per hour.
• Position feedback: Critical for imitation learning. Grippers with encoder-based position feedback (Robotiq 2F-85, OnRobot RG2) report exact jaw aperture. Binary grippers (open/close pneumatic) report only open or closed state.
• Weight: Added to the arm's end, reducing effective payload. The Robotiq 2F-85 weighs 0.9 kg; the Schunk PGN-plus-80 weighs 0.45 kg. Every gram at the end effector costs payload capacity.

Top products:

• Robotiq 2F-85 ($4,200) — the default choice in robotics research. 85 mm stroke, 235 N grip force, position feedback with 0.4 mm resolution, 0.9 kg. Plug-and-play with UR, Franka, and most major arms. UR+ certified.
• OnRobot RG2 ($2,800) — lighter (0.65 kg), 110 mm stroke, 40 N max force. Better for smaller payloads. Built-in depth sensor option. Wider opening suits more object sizes.
• Schunk PGN-plus-80 ($1,200-2,500) — industrial workhorse. Available in pneumatic and electric versions. Extremely reliable (rated 10M+ cycles). Heavier build but bulletproof for production.
• Franka Hand ($0, included with Franka arm) — purpose-built for Franka Emika Panda. 80 mm stroke, integrated force sensing via arm's torque sensors, 0.7 kg. Not compatible with other arms.

Deep Dive: Dexterous Hands

Dexterous robot hands replicate the human hand's ability to manipulate objects using multiple independently controlled fingers. They represent the frontier of end effector technology — maximum capability at maximum cost and complexity.

Why dexterous hands matter: Roughly 40% of human manipulation tasks require in-hand re-orientation, finger gaiting, or multi-contact stabilization that is physically impossible with a parallel jaw gripper. Tasks like rotating a pen, inserting a key, tying a knot, or assembling small components demand dexterity that only multi-finger hands provide.

Leading dexterous hands in 2026:

• Allegro Hand (~$15,000) — 4 fingers, 16 DOF, torque-controlled. The most widely used research hand. Compatible with ROS2 and common RL frameworks. Payload: ~1.5 kg. Weakness: no tactile sensing by default (third-party sensor skins available at $3,000-8,000 extra).
• Shadow Dexterous Hand ($150,000+) — 5 fingers, 24 DOF, the most anthropomorphic hand available. Pneumatic or electric actuation. Integrated BioTac tactile sensors ($20,000 option). Used by top research labs (OpenAI, DeepMind) for manipulation research. Extremely fragile and requires dedicated maintenance.
• Inspire Hand RH56 (~$8,000) — 5 fingers, 12 DOF. The best value in dexterous hands. Designed in China, rapidly gaining adoption in research. Electric actuators, position and current feedback. Compatible with standard ISO flanges. Less dexterous than Shadow but 18x cheaper.
• BrainCo Revo Hand (~$12,000) — 5 fingers, 11 DOF, originally developed as a prosthetic. Strong grip force (whole hand: 10 kg), lightweight (0.5 kg), integrated EMG interface. Increasingly used in teleoperation for dexterous data collection.
• ORCA Hand (open source, ~$2,000 in parts) — 4 fingers, 16 DOF. Developed at ETH Zurich, fully open-source hardware design. 3D-printable structural components with off-the-shelf Dynamixel servos. Limited payload and durability, but unbeatable for budget-constrained research and education.

For most teams starting with dexterous manipulation, the Inspire RH56 offers the best balance of capability, price, and reliability. The Allegro Hand remains the standard for research benchmarking. Shadow is only justified for labs with dedicated hand maintenance staff and >$200K hardware budgets.

Deep Dive: Vacuum Grippers

Vacuum grippers use negative air pressure to create suction against an object's surface. They dominate high-speed logistics and packaging because they can pick faster than any mechanical gripper — a suction cup engages in under 50 ms and releases instantly by venting pressure.

Bernoulli vs. suction cup: Suction cups form a physical seal against the object and pull a vacuum inside the cup. Bernoulli grippers use the Coanda effect to create a low-pressure zone beneath the gripper without touching the object. Bernoulli grippers work on porous materials (cardboard, fabric) where suction cups fail, but generate less holding force.

Suction cup types:

• Flat cups: Best for smooth, flat surfaces (glass, metal sheets, flat packaging). Highest holding force per unit area. Will not seal on curved or textured surfaces.
• Bellows cups: Accordion-style cups that conform to curved or uneven surfaces. 2-4 pleats provide 10-30 mm of height compliance. Lower holding force than flat cups but dramatically wider object compatibility.
• Oval cups: Optimized for elongated objects (tubes, rails, rectangular packages). Distribute suction along the object's long axis to prevent rotation during transport.

Key specifications: Cup diameter (determines max holding force), vacuum level (typically -0.6 to -0.85 bar), holding force (calculated as cup area x vacuum level), response time (50-200 ms for engage, near-instant for release via blow-off valve), and air consumption (higher for Bernoulli, near-zero for sealed suction cups once engaged).

Top products: OnRobot VGC10 ($3,500, electric vacuum, no external compressor needed), Schmalz FXCB ($600-2,000, configurable cup arrays), Piab piCOBOT ($2,800, integrated vacuum pump, COBOT-optimized).

Force-Torque Sensors

A force-torque (F/T) sensor mounts between the arm flange and the end effector, measuring forces and torques in all six axes (Fx, Fy, Fz, Tx, Ty, Tz). It is not an end effector itself but an enabling component that transforms any gripper into a force-aware tool.

Why they matter: Without force feedback, a robot applies blind force. Contact tasks — insertion, polishing, assembly, quality inspection by touch — require knowing how much force the end effector exerts. For imitation learning and teleoperation, F/T data adds a critical observation channel that improves policy robustness on contact-rich tasks.

Top options:

• ATI Mini45 ($5,000-8,000) — the gold standard. 6-axis, 0.01 N resolution, 2000+ Hz sample rate. Used in most research labs. Extremely accurate but expensive and requires a dedicated DAQ.
• Robotiq FT 300-S ($3,500) — designed for UR arms. 6-axis, built-in signal processing, USB output at 100 Hz. Lower resolution than ATI but dramatically simpler integration — plug into UR arm and read via URScript.
• OnRobot HEX-E/H ($3,200) — 6-axis with integrated mounting for OnRobot grippers. Optical sensing (not strain gauge) for better noise resistance in electrically noisy environments.
• Bota Systems SensONE ($2,500) — compact, lightweight (0.08 kg), EtherCAT interface. Popular in legged robotics and as a wrist sensor for manipulation research.

Matching End Effectors to Tasks

This decision matrix maps common task types and object properties to the recommended end effector type:

Decision shortcut: If you are unsure, start with a Robotiq 2F-85 parallel jaw gripper. It handles the widest range of tasks acceptably. Only switch to a specialized end effector when the parallel jaw demonstrably fails at your specific task — and it will tell you clearly when it does (drops objects, cannot grasp flat items, crushes fragile parts).

Compatibility Guide

Mechanical compatibility between arms and end effectors centers on the ISO 9409-1 flange standard. Here is what works with the most common robot arms:

Adapter plates: When a gripper does not mount directly to your arm's flange, a simple aluminum adapter plate ($50-200 CNC machined, or 3D-printed for prototyping) bridges the bolt patterns. SVRC designs and stocks adapter plates for common arm/gripper combinations.

Cost Breakdown

Teleoperation End Effectors

When a human teleoperates a robot, the end effector requirements change in important ways compared to autonomous operation:

• Compliance: Teleoperators cannot react as fast as a programmed controller. The end effector should have mechanical compliance (spring-loaded fingers, soft materials) to absorb impact forces from operator timing errors. Rigid grippers driven by an imprecise human operator will damage objects and the gripper itself.
• Position feedback: The operator needs to see or feel gripper state. End effectors with continuous position feedback (not just open/close) give the operator proportional control and better situational awareness. This is also critical for recording high-quality imitation learning demonstrations.
• Latency sensitivity: End effector actuation speed must match the teleoperation control loop. If the arm moves at 50 Hz but the gripper takes 500 ms to close, the operator experiences frustrating lag between commanding a grasp and seeing it execute. Target gripper actuation within 100 ms for responsive teleoperation.
• Simplicity of action space: For teleoperation data collection, simpler is better. A binary (open/close) or 1-DOF (aperture width) gripper requires only one control input from the operator. Dexterous hands with 16+ DOF require exoskeleton gloves to teleoperate and introduce proportionally more noise into the recorded demonstrations.

Recommendation for teleoperation data collection: Robotiq 2F-85 (binary open/close mapped to VR trigger) or Franka Hand (proportional control via SpaceMouse button) for standard manipulation. Inspire RH56 with SenseGlove Nova for dexterous manipulation tasks. Avoid Shadow Hand for teleoperation data collection unless you have a $50K+ haptic glove setup and trained operators.

Common Mistakes When Choosing End Effectors

• Buying the most expensive option: A Shadow Dexterous Hand on a task that a Robotiq 2F-85 can handle is $146,000 of waste plus months of integration time. Start with the simplest gripper that could work, and upgrade only when it demonstrably fails.
• Ignoring weight: A 2 kg gripper on a 5 kg payload arm leaves only 3 kg for the actual object. Always subtract end effector weight from the arm's rated payload. For cobots with 3-5 kg payload, this constraint is severe — a 0.9 kg Robotiq 2F-85 consumes 18-30% of available payload.
• No adapter plate planning: Teams order a gripper and discover it does not bolt onto their arm. Check the flange standard first. Order the adapter plate at the same time as the gripper. Budget 2-4 weeks for custom adapter fabrication.
• Underestimating fingertip design: Default flat fingertips on a parallel jaw gripper work for 60% of objects. Custom fingertips (V-groove for cylinders, soft-coated for fragile items, serrated for heavy parts) solve most remaining grasp failures. Budget $50-200 for custom fingertip sets — the highest-ROI upgrade on any gripper.
• Choosing vacuum for the wrong surface: Vacuum grippers fail on porous, wet, oily, dusty, or highly curved surfaces. Teams buy vacuum grippers for "general purpose" use and discover they work on only half their objects. Test with your actual objects before committing.

Frequently Asked Questions

• What is the most common robot end effector? The parallel jaw gripper accounts for over 60% of industrial robot installations. The Robotiq 2F-85 and Schunk PGN-plus are the most widely deployed models.
• What is the difference between a gripper and an end effector? A gripper is a type of end effector designed for grasping. End effector is the broader term covering any tool on a robot arm — grippers, welding torches, suction cups, cameras, and sensors are all end effectors.
• How do I attach an end effector to a robot arm? Most arms use an ISO 9409-1 standard flange. Common sizes are 50 mm (UR5e, xArm) and 31.5 mm (smaller arms). End effectors bolt directly to the flange or connect through an adapter plate.
• What end effector is best for imitation learning? A parallel jaw gripper with position feedback (Robotiq 2F-85). The simple action space reduces policy complexity, and position feedback provides ground-truth gripper state for observations.
• How much does a robot gripper cost? $200 for basic hobby-grade to $150,000+ for research-grade dexterous hands. Most research labs spend $2,000-5,000 on a mid-range parallel jaw gripper.
• Can I use a vacuum gripper on curved surfaces? Yes, with bellows-style suction cups that conform to the curvature. Flat cups require flat surfaces. The curvature must be gentle enough for the cup to maintain a seal.
• What is a tool changer and when do I need one? A tool changer lets a robot swap end effectors automatically. You need one when a single workflow requires multiple tool types. They add 30-80 mm height, reduce payload by 0.5-2 kg, and cost $2,000-15,000.
• What end effector works with a UR5e? The UR5e uses a 50 mm ISO flange. Plug-and-play options: Robotiq 2F-85/140, OnRobot RG2/RG6, Schunk Co-act. Any end effector with a 50 mm ISO adapter will physically mount.