Collaborative robot
Last reviewed
Apr 28, 2026
Sources
17 citations
Review status
Source-backed
Revision
v4 · 4,137 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Apr 28, 2026
Sources
17 citations
Review status
Source-backed
Revision
v4 · 4,137 words
Add missing citations, update stale details, or suggest a clearer explanation.
A collaborative robot (commonly called a cobot) is a robot designed to physically interact with humans in a shared workspace. Unlike traditional industrial robots that operate behind safety cages, cobots are built with sensors, force-limited joints, and rounded surfaces that allow them to work safely alongside human operators. The term "cobot" was coined at Northwestern University in 1996, and the technology has since grown into one of the fastest-expanding segments of the global robotics industry.
The formal definition of a collaborative robot comes from international safety standards governing robotic systems. ISO 10218-1 and ISO 10218-2, titled "Safety Requirements for Industrial Robots," establish the foundational safety framework for all industrial robots, including those used in collaborative applications. These standards were updated and republished as ISO 10218-1:2025 and ISO 10218-2:2025.
ISO/TS 15066:2016, titled "Robots and robotic devices: Collaborative robots," is a technical specification that supplements ISO 10218. It provides detailed guidance on implementing collaborative robot systems and defines allowable force and pressure thresholds for contact between a robot and a human body across 29 specific body regions. The central principle of ISO/TS 15066 is that if contact between a robot and a human is permitted, that contact must not result in pain or injury.
ISO/TS 15066 identifies four collaborative operation modes:
| Mode | Description | Typical Use Case |
|---|---|---|
| Safety-rated monitored stop | The robot halts all motion before a human enters the collaborative workspace. The robot resumes only after the human exits and a restart signal is issued. | Loading parts into a fixture while the robot waits |
| Hand guiding | A human operator physically moves the robot by hand using a force/torque sensor at the end effector. The robot follows the operator's guidance in real time. | Teaching new paths, precise positioning tasks |
| Speed and separation monitoring | The robot adjusts its speed based on the measured distance to a nearby human. If the human moves closer, the robot slows down or stops. Area scanners and vision systems track operator position. | Shared workstations where the operator occasionally enters the robot's zone |
| Power and force limiting | The robot's design and control system restrict the forces and pressures the robot can exert. Contact between the robot and a human is permitted but constrained to levels that do not cause injury. | Pick-and-place tasks, light assembly, polishing |
A given cobot application may use one or more of these modes simultaneously. Power and force limiting is the most common mode used in dedicated cobot hardware because it allows continuous human-robot proximity without supplementary safety devices.
The concept of the collaborative robot originated in 1995 when mechanical engineering professors J. Edward Colgate and Michael Peshkin at Northwestern University began a research project funded by a General Motors Foundation grant. Observing automobile assembly line workers as they guided bulky, awkward parts into vehicles, Colgate and Peshkin set out to build a device that could help guide payloads along constrained paths while a human provided the motive force.
Their first cobots, completed in 1996, were intentionally passive. They had no independent power source. A human co-worker supplied the force needed to move a payload, and the cobot contributed a computer-controlled guidance pathway. If the payload started moving in the wrong direction, the cobot smoothly redirected it back onto the correct path. Colgate and Peshkin initially called their invention a "programmable constraint machine." Seeking a more accessible name, they offered a $50 reward to the lab member who could suggest something better. Post-doctoral fellow Brent Gillespie (now a professor at the University of Michigan) proposed "cobot," and the name stuck.
Colgate and Peshkin published their foundational work in a 1996 paper and were later awarded U.S. patents for their cobot designs. Their work established the core principle that a cobot should complement human capabilities rather than replace them.
Universal Robots was founded in 2005 in Odense, Denmark, by Esben Ostergaard, Kasper Stoy, and Kristian Kassow. The three co-founders, who had collaborated at the University of Southern Denmark, recognized that the robotics market was dominated by heavy, expensive, and difficult-to-program machines that were impractical for small and medium-sized enterprises. In 2008, Universal Robots released the UR5, a six-axis articulated robot arm with a 5 kg payload, widely regarded as the first commercially viable collaborative robot. The UR5 could operate without safety fencing, was lightweight enough for a single worker to install, and could be programmed through an intuitive touchscreen interface.
The success of Universal Robots spurred other major robotics companies to develop their own collaborative product lines. KUKA introduced the LBR iiwa in 2013, ABB launched its YuMi dual-arm cobot in 2015, and FANUC released its CR series around the same time. In 2015, Teradyne acquired Universal Robots for $285 million (with up to $65 million in additional performance-based payments), validating the commercial potential of collaborative robotics.
Universal Robots is the most commercially successful collaborative robot manufacturer in the world. The company has sold over 100,000 cobots worldwide as of 2025, capturing an estimated 40 to 50 percent of the global cobot market share. Universal Robots offers three product series.
The e-Series, introduced in 2018, features built-in force/torque sensing at the tool flange and serves as the company's workhorse line for small and medium payloads. In 2025, Universal Robots renamed and upgraded two e-Series models: the UR5e became the UR7e (payload increased to 7 kg, 1300 mm reach, repeatability of plus or minus 0.03 mm), and the UR10e became the UR12e (12 kg payload, 1300 mm reach, repeatability of plus or minus 0.05 mm). The series also includes the UR3e (3 kg payload, 500 mm reach) and the UR16e (16 kg payload, 900 mm reach).
The UR Series represents the company's flagship high-performance line, designed for applications demanding higher payloads and longer reaches:
The original UR5 (2008), UR10 (2012), and UR3 (2015) are no longer in production but remain widely deployed.
In October 2024, Universal Robots introduced the UR AI Accelerator, a hardware and software toolkit integrating an NVIDIA Jetson AGX Orin module, an Orbbec Gemini 335Lg 3D camera, and PolyScope X with a native ROS 2 interface. The toolkit enables developers to build AI-powered cobot applications with capabilities such as object detection, pose estimation, path planning, image classification, and quality inspection.
Several established robotics companies and newer entrants compete in the collaborative robot market.
FANUC, the world's largest industrial robot maker by installed base, offers the CRX collaborative robot series. The lineup spans payloads from 5 kg to 30 kg:
All CRX models feature IP67 dust and water protection, FANUC's R-30iB Mini Plus controller, drag-and-drop tablet programming, and an advertised maintenance-free lifespan of eight years.
ABB offers two collaborative robot families:
GoFa (CRB 15000) is available in three payload variants: GoFa 5 (5 kg payload, 950 mm reach), GoFa 10 (10 kg payload, 1520 mm reach), and GoFa 12 (12 kg payload, 1370 mm reach). All GoFa models feature torque sensors in all six joints, a maximum speed of 2.2 m/s, and best-in-class repeatability of plus or minus 0.02 mm. They are powered by ABB's OmniCore controller.
SWIFTI (CRB 1100) is designed for tasks that require higher speeds while maintaining collaborative safety. It supports a 4 kg payload with up to 580 mm reach and can move at speeds up to 5 m/s. SWIFTI achieves collaborative safety by combining ABB's SafeMove software with a safety laser scanner for speed and separation monitoring rather than power and force limiting.
KUKA introduced the LBR iiwa ("intelligent industrial work assistant") in 2013 as one of the first purpose-built collaborative robots from a major industrial robotics company. It is a seven-axis robot arm available in two versions:
The LBR iiwa's seven axes give it human-arm-like flexibility. Joint torque sensors in all axes enable immediate contact detection with a pose repeatability of plus or minus 0.1 mm. The robot meets Category 3, Performance Level d safety standards per EN ISO 13849-1.
Doosan Robotics, founded in 2015 and headquartered in Suwon, South Korea, went public on the Korea Exchange in October 2023 in what became South Korea's largest IPO of that year, raising approximately 421 billion won ($317 million). The company offers the M-Series of collaborative robots:
Doosan cobots feature six-axis torque sensors with a touch sensitivity of 0.2 Nm and repeatability ranging from plus or minus 0.03 to 0.1 mm depending on the model. The company also offers the H-Series for heavier payloads up to 25 kg and the A-Series for shorter-reach, high-precision applications.
Techman Robot, a subsidiary of Quanta Computer, is a Taiwanese manufacturer known for integrating built-in computer vision directly into its cobot hardware. Techman's cobots ship with TMvision, a vision system that supports barcode reading, image-based alignment, optical character recognition, and one-click calibration without requiring external cameras. Programming is done through TMflow, a drag-and-drop visual programming interface. Techman's product line ranges from 4 kg to 25 kg payload with reaches from 700 mm to 1900 mm. All Techman cobots comply with ISO 10218-1 and ISO/TS 15066.
Franka Emika, founded in Munich, Germany, by Sami Haddadin, gained recognition for the Panda, a seven-axis cobot with a 3 kg payload, 850 mm reach, and 0.1 mm repeatability. Weighing just 18 kg, the Panda was widely adopted in research labs and universities due to its open-source Franka Control Interface (FCI) operating at 1 kHz for real-time torque control. Franka Emika filed for insolvency in August 2023 due to shareholder disputes. In November 2023, Agile Robots AG acquired the company's operations and approximately 100 employees for a reported sum exceeding 30 million euros. The company now operates as Franka Robotics under the Agile Robots umbrella.
| Manufacturer | Model | Axes | Payload (kg) | Reach (mm) | Repeatability (mm) | Weight (kg) | Key Feature |
|---|---|---|---|---|---|---|---|
| Universal Robots | UR7e | 6 | 7 | 1300 | +/-0.03 | 20.6 | Upgraded e-Series, built-in F/T sensor |
| Universal Robots | UR20 | 6 | 20 | 1750 | +/-0.05 | 64 | Long reach, high payload |
| Universal Robots | UR30 | 6 | 30 | 1300 | +/-0.05 | 63.5 | Heaviest UR payload |
| FANUC | CRX-10iA | 6 | 10 | 1249 | +/-0.03 | 40 | IP67, 8-year maintenance-free |
| FANUC | CRX-25iA | 6 | 25 | 1889 | +/-0.03 | 67 | Longest CRX reach |
| ABB | GoFa 12 | 6 | 12 | 1370 | +/-0.02 | ~35 | Best-in-class repeatability |
| ABB | SWIFTI CRB 1100 | 6 | 4 | 580 | +/-0.01 | ~25 | 5 m/s speed, laser scanner safety |
| KUKA | LBR iiwa 14 | 7 | 14 | 820 | +/-0.1 | 30 | 7-axis, human-arm flexibility |
| Doosan Robotics | M1013 | 6 | 10 | 1300 | +/-0.05 | ~33 | 0.2 Nm touch sensitivity |
| Techman Robot | TM12 | 6 | 12 | 1300 | +/-0.05 | ~33.3 | Built-in vision system |
| Franka Robotics | Panda / FR3 | 7 | 3 | 850 | +/-0.1 | 18 | 1 kHz real-time torque control |
Collaborative robots are deployed across a wide range of industries and use cases. Their flexibility, ease of programming, and ability to share workspaces with humans make them particularly well-suited for tasks that are repetitive, ergonomically challenging, or require frequent changeover.
Machine tending is one of the most common cobot applications. Cobots load raw materials into CNC machines, injection molding presses, 3D printers, or stamping machines, then unload finished parts when the cycle completes. The cobot handles the repetitive loading and unloading, while a human operator oversees quality and manages exceptions. This arrangement allows a single operator to supervise multiple machines simultaneously.
Cobots perform assembly tasks including screw driving, snap fitting, part insertion, and component placement. In automotive manufacturing, cobots assist workers with installing trim pieces, fastening bolts, and applying adhesives. In electronics manufacturing, cobots handle delicate components such as circuit boards and connectors. The ability to program a cobot by physically guiding its arm along the desired path (lead-through teaching) makes it straightforward to set up new assembly sequences without writing code.
Equipped with cameras and vision systems, cobots perform visual quality inspection by examining parts for surface defects, dimensional accuracy, and correct assembly. The cobot positions a camera or sensor at precise locations around a part, captures images, and passes them to machine learning algorithms for analysis. This approach provides consistent, repeatable inspection without the fatigue that affects human inspectors on long shifts.
Cobots handle primary packaging (placing products into boxes), secondary packaging (grouping boxes into cases), and palletizing (stacking cases onto pallets). Higher-payload cobots such as the UR20, UR30, and FANUC CRX-25iA are particularly suited for palletizing applications. Cobots can be quickly reprogrammed when product sizes or packaging configurations change, giving them a significant advantage over dedicated palletizing machinery in facilities that handle many different SKUs.
In pharmaceutical, biotech, and chemical research laboratories, cobots automate sample handling, pipetting, plate loading, and instrument tending. Cobots reduce the risk of human contamination in controlled environments and improve throughput for repetitive lab procedures. Their compact footprint allows them to be placed inside fume hoods or biosafety cabinets.
Collaborative welding has become a growing application, particularly for small-batch and custom fabrication shops that cannot justify the cost of a traditional robotic welding cell. A welder programs the cobot by guiding it through the weld path by hand, and the cobot then repeats the path with consistent speed and torch angle. Products like the Hirebotics Beacon system, built on Universal Robots hardware, simplify cobot welding setup through smartphone-based programming.
Additional cobot applications include surface finishing (sanding, polishing, deburring), gluing and dispensing, testing and measurement, material handling, and pick-and-place operations.
Safety is the defining characteristic that separates collaborative robots from traditional industrial robots. While a conventional industrial robot operates at high speeds behind physical barriers, a cobot is engineered to be inherently safe during close interaction with humans.
Collaborative robots incorporate several design features to minimize injury risk:
ISO/TS 15066 does not automatically make any cobot application safe. Every collaborative robot installation requires a thorough risk assessment that considers the specific robot, end effector, workpiece, task, and workspace layout. A cobot with a sharp tool or a heavy workpiece may require additional safeguards even though the robot itself is force-limited. The risk assessment determines which collaborative mode (or combination of modes) is appropriate and sets the specific force and speed thresholds for the application.
The collaborative robot market has experienced rapid growth since Universal Robots commercialized the category in 2008. According to the International Federation of Robotics (IFR), cobots accounted for 10.5 percent of all industrial robot installations worldwide in 2023, up from roughly 3 percent in 2017.
Market size estimates vary by research firm due to differences in methodology and scope:
| Source | 2025 Market Size (USD) | 2030 Forecast (USD) | CAGR |
|---|---|---|---|
| MarketsandMarkets | $1.42 billion | $3.38 billion | 18.9% |
| Grand View Research | $2.95 billion | ~$17.2 billion (by 2033) | 23.1% |
| ABI Research | N/A | $7 billion | 27.5% |
| Research and Markets | $4.18 billion | $15.55 billion | 30.1% |
Despite variance in the exact figures, all major research firms agree that the cobot market is growing at a compound annual growth rate (CAGR) of roughly 19 to 30 percent, driven by labor shortages in manufacturing, declining cobot prices, and expanding applications beyond traditional factory settings.
According to the IFR's World Robotics 2025 report, approximately 542,000 industrial robots were installed worldwide in 2024, marking the fourth consecutive year that annual installations exceeded 500,000 units. Asia accounted for 74 percent of all new deployments.
The integration of artificial intelligence is expanding what collaborative robots can do and simplifying how they are programmed.
Computer vision systems, powered by deep learning models, give cobots the ability to identify, locate, and classify objects in unstructured environments. Vision-equipped cobots can pick randomly oriented parts from bins (bin picking), inspect products for defects, and adapt to variations in part placement. Techman Robot integrates a vision system directly into the cobot wrist, while Universal Robots and FANUC offer vision add-ons through their partner ecosystems. The UR AI Accelerator's combination of an NVIDIA Jetson AGX Orin and a 3D depth camera represents a purpose-built approach to enabling AI vision on cobots.
Advanced force and torque sensors enable cobots to perform tasks that require a delicate touch, such as inserting tightly fitting parts, polishing curved surfaces, or testing the click force of buttons. AI algorithms process force data in real time to adjust the robot's behavior, combining data from multiple sensors (sensor fusion) to achieve precise force application. The KUKA LBR iiwa and Franka Panda, with joint torque sensors in all seven axes, have been particularly influential in research on force-controlled manipulation.
Programming by demonstration (also called learning from demonstration or lead-through teaching) allows a non-expert to teach a cobot a new task by physically guiding the robot arm through the desired motions. The cobot records the trajectory and can replay it autonomously. AI-enhanced versions of this approach use machine learning to generalize from a small number of demonstrations, enabling the cobot to adapt to variations in part position or orientation. Research labs are also exploring mixed-reality teleoperation, where an operator uses haptic devices and AR/VR headsets to demonstrate tasks remotely.
Emerging research explores using large language models and voice commands to instruct cobots. Rather than programming waypoints or writing code, an operator could describe a task in plain language, and the cobot would translate the instructions into a sequence of actions. While still largely in the research stage as of 2025, natural language interfaces represent a potential step change in cobot accessibility.
Modern cobots increasingly use edge computing for real-time inference (running AI models locally on hardware like the NVIDIA Jetson) while connecting to cloud platforms for fleet management, data analysis, and model updates. This hybrid approach enables cobots to react quickly to their immediate environment while benefiting from centralized data aggregation across multiple robots and sites.
Collaborative robots and traditional industrial robots serve different but overlapping roles in manufacturing and automation.
| Feature | Collaborative Robot | Traditional Industrial Robot |
|---|---|---|
| Safety approach | Force-limited, can share workspace with humans | Requires safety fencing or barriers |
| Typical payload | 3 to 35 kg | Up to 1,000+ kg |
| Typical speed | 1 to 5 m/s (TCP) | Up to 10+ m/s |
| Programming | Lead-through teaching, drag-and-drop, tablet interfaces | Teach pendant, offline programming, specialized languages |
| Setup time | Hours to days | Days to weeks |
| Footprint | Compact, can be mounted on tables or mobile platforms | Large, requires dedicated floor space and fencing |
| Cost | $25,000 to $85,000 (robot arm only) | $50,000 to $300,000+ (robot arm only) |
| ROI timeline | Often under one year | Typically two to four years |
| Best suited for | Small to medium batch sizes, mixed-product lines, tasks requiring human proximity | High-volume, high-speed production, heavy payloads, hazardous environments |
Many manufacturers deploy both types. Traditional industrial robots handle the high-speed, high-payload core of a production line, while cobots manage flexible, lower-volume tasks at the edges. The boundary between the two categories is blurring as cobots grow more capable and industrial robots add collaborative modes.
Despite rapid growth, collaborative robots face several challenges: