Collaborative Robots

Buying guide

Overview

Collaborative robots (cobots) are robot arms designed to work directly alongside humans on a shared workspace without safety caging. Unlike traditional industrial robots that operate behind fences, cobots use force-torque sensing, vision, and compliant motion to detect and react to human presence — stopping or slowing before causing injury.

They are the entry point for automation in small-to-medium factories, assembly lines, labs, and warehouses that lack the space, budget, or volume for traditional industrial robots.

Market Snapshot:

Global cobot market valued at ~$1.5B in 2024, projected to reach ~$10B by 2030 (CAGR ~30%)
Top brands: Universal Robots (UR), FANUC CRX, ABB GoFa/SWIFTI, KUKA LBR iisy, Techman Robot, Doosan, Elephant Robotics, Franka Emika
Purchase price range: $15,000–$80,000/unit (arm only); $25,000–$150,000 fully integrated
Cobot-as-a-Service (CaaS): $1,500–$4,000/month

Buyer Personas

Persona	Primary Pain Point	What They're Buying
SME Factory Owner / Operations Manager	Labor shortages, rising wages, repetitive injury claims	A machine that works the night shift without complaining
Manufacturing Engineer / Automation Lead	Need to justify ROI, prove cycle time improvement	Reliable specs, integration docs, proven uptime
Quality / Lab Manager (pharma, electronics)	Precision, repeatability, audit trail	Sub-millimeter accuracy, ISO/cleanroom compliance
Systems Integrator / VAR	Fast deployment, broad end-effector ecosystem	Open APIs, ROS support, strong local distribution
Procurement / CFO	Total cost of ownership, not sticker price	Leasing options, maintenance SLA, payback period

Spec Reference

Arm Mechanics — "What Can It Physically Do?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`payload_kg`	Maximum weight it can hold at the tip of the arm	3–35 kg	5–16 kg	The single most important number. A 5 kg cobot cannot pick up a car door panel. Always size up — payload drops as reach extends.
`reach_mm`	How far the arm can stretch from its base	500–1800 mm	900–1300 mm	Determines how large a work cell it can serve. A short reach means moving the robot or the workpiece more often.
`degrees_of_freedom`	How many joints it has (= how flexibly it can move)	4–7 DOF	6 DOF	6 DOF mimics a human arm. 7 DOF adds a "wrist twist" for tight spaces. 4 DOF is cheaper but limited to flat pick-and-place only.
`repeatability_mm`	How precisely it returns to the same point, every time	±0.01–±0.10 mm	±0.03–±0.05 mm	At ±0.1 mm, it cannot reliably insert a USB connector. At ±0.02 mm, it can. Critical for assembly, dispensing, and electronics.
`max_speed_deg_s`	How fast the joints can rotate	90–360°/s	180–225°/s	Faster = shorter cycle time = more parts per hour. But faster cobots near humans require more safety validation.
`robot_weight_kg`	How heavy the arm itself is	9–78 kg	18–33 kg	Lighter arms are easier to re-deploy to a new station. Heavy arms need a structural mount — you can't bolt them to a folding table.
`mounting_options`	What orientations it can be installed in	Floor, ceiling, wall, tilted	Any-angle (universal)	Ceiling-mounted cobots free up floor space. Not all models support inverted mounting — check before designing the cell.
`ip_rating`	Dust and water resistance	IP40–IP67	IP54 (standard); IP67 (food/pharma)	In food production or wet environments, water ingress will destroy unrated electronics within weeks.

Safety — "Will It Hurt Someone?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`safety_standard`	Official certifications it has passed	ISO 10218, ISO/TS 15066, EN ISO 13849	ISO/TS 15066 (the cobot-specific standard)	Without ISO/TS 15066 compliance, your insurance won't cover a workplace injury involving this robot.
`force_torque_sensing`	Can it feel if it pushes against a human?	Joint-level vs. External sensor vs. None	Joint-level (built-in, every joint)	A cobot without force sensing is just a fast robot with no cage — dangerously misleading. Joint-level sensing catches collisions anywhere on the arm.
`collision_detection_sensitivity`	How gentle a "bump" triggers a stop	Adjustable 0–100%	Adjustable (10 sensitivity levels)	In delicate assembly, you want it to stop at the slightest resistance. In heavy grinding, false stops kill productivity. You need to tune this.
`power_force_limiting` (PFL)	Maximum push force before it stops	50–150 N	100–150 N at full stop	ISO/TS 15066 defines maximum contact forces for different body parts. Vendor must prove their robot meets these limits at your intended speed.
`tcp_speed_at_reduced_mode_ms`	How fast it moves when a human is nearby	0.1–0.5 m/s	0.25 m/s (SSM trigger)	Speed and Separation Monitoring (SSM) slows the robot as you approach. Too slow = workers hate it and defeat the safety sensor.
`safety_io_ports`	Dedicated safety signal inputs/outputs	2–8 safety I/O pairs	≥ 4 safety I/O pairs	Needed to connect safety scanners, light curtains, and emergency stops that meet SIL2/PLd requirements.
`safety_category_rating`	How fail-safe the safety system is	Cat.2 / PLc — Cat.3 / PLd	Cat.3 PLd (minimum for human collaboration)	PLd means 2 independent hardware faults must occur simultaneously before a dangerous failure. PLc has single-point failure risk.

Ease of Use — "Can My Team Actually Run This?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`programming_method`	How you tell it what to do	Teach pendant, graphical (drag-and-drop), hand-guiding, code (URScript, Python)	Hand-guiding + graphical UI	Hand-guiding = grab the arm and physically move it through the path. No coding required. Critical for SMEs without automation engineers.
`setup_time_hours`	Hours from box to first cycle	2–40 hours	< 8 hours	A cobot that takes 3 weeks to set up is not agile. The best cobots run a pick-and-place demo in an afternoon.
`no_code_capable`	Can a non-programmer deploy it?	Yes / No / Partial	Yes (for standard tasks)	If your only programmer leaves, can your line supervisor keep the robot running? This is a business continuity question.
`teach_pendant_display`	Screen size and type on the handheld controller	7"–12" touchscreen	10"+ HD touchscreen	Small, low-resolution screens slow down every setup and troubleshooting session.
`offline_programming_support`	Can you simulate and program it in software before it arrives?	None / Proprietary / ROS/RoboDK	RoboDK or proprietary simulator included	Offline programming means zero production downtime for changeovers. Essential for high-mix, low-volume manufacturing.
`time_to_redeploy_min`	How long to move it to a different task	15–480 min	< 60 min	The entire cobot value proposition for SMEs is flexibility. If redeployment takes a full day, it's not flexible.

End-Effector & Tooling — "What Can I Attach to It?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`tool_flange_standard`	The mechanical connector at the tip of the arm	ISO 9283 (various sizes), proprietary	ISO 9283 compliant	Non-standard flanges lock you into the vendor's own grippers. ISO-compliant flanges let you use Robotiq, Schunk, OnRobot, and 200+ others.
`tool_changer_support`	Can it automatically swap end-effectors mid-task?	None / Manual / Automatic	Automatic (for multi-task cells)	Without a tool changer, a cobot that needs to pick then weld requires a full stop and manual swap. Automatic changers keep the line moving.
`end_effector_power_supply`	Electrical and pneumatic connections at the wrist	12V/24V DC, pneumatic (6 bar)	24V DC + pneumatic passthrough	If the wrist has no power passthrough, you're zip-tying cables down the arm. Messy, breaks fast, catches on things.
`integrated_force_torque_sensor`	A wrist-mounted sensor for delicate insertion tasks	None / Optional / Built-in	Built-in or available as option	Without F/T sensing at the wrist, tasks like inserting a peg into a hole require ±0.01 mm precision in positioning. With F/T, you can feel your way in.

Connectivity & Software — "Does It Fit Into My Systems?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`communication_protocols`	Industrial languages it speaks	Modbus TCP, Profinet, EtherNet/IP, EtherCAT, OPC-UA	OPC-UA + EtherNet/IP minimum	If your PLC speaks Profinet and the cobot only speaks Modbus, you need an expensive gateway. Check your existing infrastructure first.
`ros_support`	Compatible with the Robot Operating System	None / ROS1 / ROS2	ROS2 supported	ROS2 is the open-source standard for robotics research and custom integration. Without it, advanced vision and AI tasks require vendor-proprietary solutions.
`sdk_language_support`	What coding languages you can use to control it	URScript, Python, C++, .NET	Python + C++ SDK	Python support means your data science team can write robot programs. Vendor-only scripting languages create dependency and hiring risk.
`open_api`	Can third parties build on top of it?	Closed / Partially open / Fully open	Open REST or gRPC API	Closed platforms mean every custom integration is a negotiation with the vendor. Open APIs let you plug in vision systems, MES, ERP.
`plc_integration`	Works with standard factory controllers	None / Via gateway / Native	Native EtherNet/IP or Profinet	Most factories already have Siemens or Allen-Bradley PLCs. The cobot must speak their language natively or you add weeks of integration work.
`digital_twin_support`	A virtual copy of the robot for simulation	None / Proprietary / Universal (RoboDK, Isaac Sim)	RoboDK + vendor simulator	Digital twins let you test a new program on a virtual robot while the real one keeps running production. Avoids costly downtime for program testing.

Performance & Reliability — "Will It Keep Running?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`mtbf_hours`	Mean Time Between Failures — expected uptime before a breakdown	20,000–60,000 hours	≥ 35,000 hours	At 8h/day, 35,000 hours = ~12 years before expected maintenance. Lower MTBF = more downtime = ROI blown.
`duty_cycle`	Can it run 24/7 or does it need rest?	8h/day rated — 24/7 continuous	24/7 continuous	If it's only rated for 1 shift, you cannot use it for overnight unmanned production. Always verify with the vendor.
`path_accuracy_mm`	How closely it follows a programmed curved path	±0.1–±1.0 mm	±0.2 mm	Critical for welding, dispensing, and painting. Poor path accuracy = bad welds, uneven glue beads, wasted material.
`max_tcp_speed_ms`	Maximum speed at the tool tip	1.0–3.5 m/s	2.0–3.0 m/s	Directly determines cycle time. At 1.0 m/s, a long-reach move takes twice as long as at 2.0 m/s. Multiplied across millions of cycles, this is your ROI.
`joint_position_resolution_deg`	Smallest joint movement it can make	0.01°–0.1°	≤ 0.02°	Fine resolution = smooth curved paths and precise positioning. Coarse resolution produces jerky motion that stresses mechanical joints.
`temperature_range_c`	Operating temperature range	0°C–45°C (standard); -10°C–55°C (extended)	0°C–45°C (standard environments)	Cold storage or foundry environments need extended ratings. Standard cobots will fault out below 0°C or above 45°C.

TCO & Commercial — "What Does It Really Cost?"

Spec	Plain English	Industry Range	Sweet Spot	Why It Matters to Buyer
`price_usd`	Arm-only purchase price	$15,000–$80,000	$25,000–$45,000	Arm-only price is misleading. Budget 2–3x for gripper, vision, integration, safety assessment, and commissioning.
`caas_monthly_usd`	Cobot-as-a-Service monthly rate	$1,500–$4,000/mo	~$2,500/mo (includes maintenance)	CaaS shifts CapEx to OpEx and typically includes software updates, preventive maintenance, and replacement units. Better for cash-constrained SMEs.
`warranty_years`	Standard repair/replace guarantee	1–3 years	2 years + optional extension	Cobots have gearboxes that wear. A 1-year warranty on a machine you plan to run for 10 years is inadequate. Ask for extended warranty pricing upfront.
`local_service_network`	Number of certified service technicians in your region	Varies widely	≥ 5 certified partners within 200 km	A cobot down for 2 weeks waiting for a technician from overseas can cost more in lost production than the robot is worth.
`spare_parts_lead_time_days`	How long to get replacement joint modules	1–30 days	< 7 days	Joint failures are the most common repair. If a replacement gearbox takes 3 weeks to arrive from Asia, your line is dead.
`typical_roi_months`	Community-reported payback period	12–36 months	18–24 months	Rule of thumb: a cobot replacing 1.5 human shifts at $25/hr pays back in ~18 months. Verify with vendor case studies in your specific application.

Hidden Concerns

3.1 The Payload-at-Reach Trap

Rated payload (e.g., 10 kg) is measured at minimum reach
At full arm extension, effective payload can drop 30–50%
Ask vendor: "What is the actual payload capacity at 80% of maximum reach for my specific task geometry?"

3.2 The Risk Assessment Obligation

ISO/TS 15066 mandates a site-specific risk assessment before deployment — the cobot certification alone is not sufficient
This assessment must be performed or validated by a certified safety engineer
Skipping it voids the cobot's collaborative rating and your liability insurance
Ask vendor: "Do you provide a risk assessment template or a certified partner to conduct the initial risk assessment at my site?"

3.3 Repeatability vs. Accuracy Confusion

Repeatability = returns to the same point consistently (what the spec sheet shows)
Accuracy = actually reaches the correct absolute position in space
Cobots have excellent repeatability but poor absolute accuracy; if you program offline and transfer to the physical arm, positions will drift by millimeters
Ask vendor: "What is the absolute positioning accuracy (not just repeatability) and do you support hand-eye calibration for vision-guided programs?"

3.4 Gearbox Backlash and Wear

Cobot joints use harmonic drive gearboxes that develop backlash (slop) over time
At 20,000+ hours, repeatability degrades from ±0.03 mm to ±0.15 mm — often without any error alert
Ask vendor: "What is the joint backlash specification at end-of-recommended-service-life, and how does the robot notify the operator when recalibration is required?"

3.5 Force-Limiting vs. Truly Safe

"Force-limiting" cobot ≠ automatically safe at all speeds
A 10 kg arm moving at 2 m/s carries significant momentum — it can break a finger even when force-limited, if the pinch geometry is wrong
Safety validation must be done at the actual deployed speed and task geometry
Ask vendor: "Can you provide the biomechanical risk calculation (per ISO/TS 15066 Annex A) for my specific application speed and payload?"

3.6 The End-Effector Ecosystem Lock-In

Some vendors use proprietary tool flanges or communication protocols that restrict which grippers work
Switching a task from suction to fingers can cost $8,000+ in adapters if the ecosystem is closed
Ask vendor: "Is the tool flange ISO 9283 compliant, and is the tool I/O wiring documented publicly so I can integrate third-party end-effectors without your involvement?"

3.7 Software Subscription Creep

Many cobots now have core features (vision, force control, advanced motion) locked behind annual software subscriptions of $2,000–$8,000/year
A $30,000 robot with $5,000/year mandatory software fees costs $80,000 over 10 years
Ask vendor: "Which features require ongoing software subscriptions, what are the annual costs, and what happens to the robot's capability if we stop paying?"

3.8 The Redeployment Myth

Vendors sell cobots as "easily redeployable" but in practice, every new task requires a new risk assessment, new program, potentially new end-effector, and retraining
True redeployment time for a non-trivial task change is typically 2–5 days, not the "30 minutes" in the brochure
Ask vendor: "Can you show me a video of your team redeploying this cobot to a completely different task from scratch — from 0 to running production?"

3.9 Collaborative Speed Limits Kill Productivity

In true collaborative mode (human in cell), ISO/TS 15066 typically limits TCP speed to 0.25–0.5 m/s
This may be 5–10x slower than the robot's rated maximum speed
Many "collaborative" deployments end up behind a light curtain running at full speed because collaborative speed is too slow for the production rate
Ask vendor: "What is the achievable cycle time for my task in full collaborative mode (no fence), and what is it with a light curtain?"

3.10 Integration Cost is Usually Larger Than Robot Cost

The arm is typically 30–40% of total system cost
Remaining cost: gripper ($3k–$15k), vision system ($5k–$20k), safety assessment ($2k–$8k), programming ($5k–$20k), fixture design ($5k–$15k), commissioning ($3k–$10k)
Ask vendor: "Can you provide a turnkey system quote, not just the arm price, for my specific application?"

How to Evaluate a Robot

A cobot must meet all criteria below:

Safety Minimums

ISO/TS 15066 certified (not just ISO 10218)
Built-in joint-level force-torque sensing on all joints
Safety category Cat.3 PLd or higher
Adjustable collision sensitivity (minimum 5 levels)
≥ 4 dedicated safety I/O pairs

Performance Minimums

Payload ≥ 5 kg
Reach ≥ 850 mm
Repeatability ≤ ±0.05 mm
Max TCP speed ≥ 1.5 m/s
MTBF ≥ 30,000 hours
24/7 duty cycle rated

Usability Minimums

Hand-guiding / lead-through programming supported
Setup time < 8 hours (documented)
ISO 9283 compliant tool flange
Offline programming software included or available

Integration Minimums

OPC-UA or EtherNet/IP native support
Python or ROS2 SDK publicly available
Open REST/gRPC API documented

Commercial Minimums

Warranty ≥ 2 years
Spare parts lead time < 10 days (documented SLA)
≥ 3 certified service partners in APAC, EMEA, and Americas respectively

Top Products Compared

Feature	UR10e	FANUC CRX-10iA	ABB GoFa CRB 15000	KUKA LBR iisy 11	Techman TM12	Doosan M1013
Payload	12.5 kg	10 kg	5 kg	11 kg	12 kg	13 kg
Reach	1300 mm	1418 mm	950 mm	1150 mm	1300 mm	1300 mm
Repeatability	±0.05 mm	±0.02 mm	±0.02 mm	±0.05 mm	±0.05 mm	±0.05 mm
Max TCP Speed	1.0 m/s	2.0 m/s	2.2 m/s	1.5 m/s	1.8 m/s	1.5 m/s
DOF	6	6	6	7	6	6
Built-in F/T Sensing	All joints	All joints	All joints	All joints	All joints	All joints
Safety Rating	Cat.3 PLd	Cat.3 PLd	Cat.3 PLd	Cat.3 PLd	Cat.3 PLd	Cat.3 PLd
Programming	Polyscope (graphical + UR Script)	iPendant + no-code tablet	Wizard Easy Programming	KUKA smartPAD	TMflow (graphical)	DART Platform
Built-in Vision	Optional	Optional	Optional	Optional	Yes (integrated)	Optional
ROS2 Support	Yes	Yes	Yes	Yes	Yes	Yes
IP Rating	IP54	IP67	IP54	IP54	IP54	IP54
Weight	33.5 kg	26 kg	24 kg	26.3 kg	33 kg	33.7 kg
Est. Price (arm only)	~$35k	~$40k	~$32k	~$45k	~$28k	~$30k
Key Differentiator	Largest installed base + ecosystem	Highest MTBF (100k hrs)	Highest speed at payload	7-DOF for tight spaces	Integrated vision camera	Best price/payload ratio

Regulations & Compliance

Regulation	Scope	What It Means for Deployment
ISO 10218-1 / -2	Robot manufacturer + integrator requirements	-1 is the robot itself; -2 is the installation. Both must be complied with.
ISO/TS 15066	Human-robot collaboration specifically	Defines maximum contact forces by body region. The mandatory standard for truly cageless operation.
EN ISO 13849-1 (PLd/Cat.3)	Safety control system design	Safety I/O and emergency stop circuits must achieve PLd minimum for cobot applications.
IEC 62061 (SIL 2)	Alternative safety integrity standard	Equivalent to PLd; some vendors certify to this instead. Both are acceptable.
OSHA 29 CFR 1910.217	US machine guarding regulations	OSHA defers to ANSI/RIA R15.06 which harmonizes with ISO 10218. Required in US workplaces.
EU Machinery Directive 2006/42/EC	CE marking requirement for EU market	CE mark mandatory. Requires a Technical Construction File and Declaration of Conformity. Transitioning to EU Machinery Regulation 2023/1230.
GDPR / CCPA	Vision system data privacy	If the cobot's vision system captures identifiable worker footage for quality control, GDPR/CCPA obligations apply.
ISO 9283	Tool flange dimensions	Standardizes mechanical interface. Compliance enables third-party end-effector use.
FDA 21 CFR Part 11	Pharma/medtech electronic records	If cobot operates in regulated pharma environment, all motion logs and process records must be audit-trail compliant.

References

ISO/TS 15066:2016 — Robots and robotic devices: Collaborative robots
IFR World Robotics 2024 Report
Universal Robots UR10e Technical Specification Sheet
FANUC CRX-10iA Product Datasheet
ABB GoFa CRB 15000 Technical Reference