How much force can a robot safely apply to a person?

There is no single answer, and this page does not provide one. Research measures where pain begins, and those thresholds vary about threefold by body location, differ between people, and differ between studies by more than double for the same location. Permissible limits for a real application are set by the robot-safety standards and a risk assessment, not by a research figure.

Why do the studies disagree?

They apply force differently, through blunt or semi-sharp surfaces, in impact or pinching, at different impactor shapes and speeds, and they test different populations. A review comparing two of them notes the contact conditions and premises differ, so the results are not directly comparable.

Which body location is most sensitive?

In the 15-site study, the arm nerve had the lowest pressure pain threshold at 65.1 N/cm2 and the back of the hand the highest at 196.1 N/cm2, roughly three times higher. Where contact happens matters more than a single force figure suggests.

Is force the only thing that matters?

No. NIST measured about 220 ms between a robot being at a position and reporting it, which matters for any approach that keeps the robot away from people rather than limiting contact. Timing and sensing lag sit alongside force in real systems.

Where are the actual limits defined?

In the robot-safety standards, principally ISO 10218-1 and ISO 10218-2, with the collaborative material formerly published as ISO/TS 15066. We do not restate their values. Implementing power and force limiting requires the standards themselves plus a risk assessment for the specific application.

Robot safety

Robot contact and human pain thresholds: what research has measured

Human pain thresholds under robot contact are not one number. In one study of 15 body sites, the pressure threshold ranged from 196.1 N/cm2 at the back of the hand down to 65.1 N/cm2 at an arm nerve, a threefold spread depending on where a person is struck.

The studies also disagree with each other. Measuring force at the deltoid, one group reported a threshold of 272 N while another, testing 112 subjects across 28 body locations, reported substantially lower values. Two careful teams, the same body part, results that differ by more than double.

This page collects what the open, peer-reviewed literature has actually measured. It does not restate any standard's limits, and it is not a specification for setting up a collaborative cell.

Data covers Peer-reviewed biomechanical studies of human pain thresholds under robot contact, 2015 to 2024. Edited by Mike Ramsey / Reliable Media.

The figures and where they come from

Each figure is rated for how safely you can cite it today. Ratings judge current usability, not whether a number was ever correct.

Figure	What it is	Source	Citation Confidence	Notes
196.1 N/cm2 (back of hand)	Highest pressure pain threshold measured	[B]	Medium	The least sensitive of 15 sites tested. Reported with a standard deviation of plus or minus 85.8, which is itself larger than the gap between many sites.
65.1 N/cm2 (arm nerve)	Lowest pressure pain threshold measured	[B]	Medium	The most sensitive site tested, roughly a third of the back of the hand. Where contact happens matters more than most discussion of cobot safety allows.
15 sites	Body sites tested	[B]	Medium	Thresholds measured three times at each site, including the forehead.
112 subjects	Subjects in the Behrens study	[A]	High	One of the larger human-subject studies in this area. Scale matters because between-person variation is wide.
28 locations	Body locations tested	[A]	High	Impact and pinching loads through blunt and semi-sharp surfaces, with force increased until the subject reported pain.
272 N	Deltoid force threshold, Han et al. (2018)	[C]	Medium	As collected in a later review comparing studies. The same review reports substantially lower deltoid values from Behrens et al., which is the disagreement this page is about.
311 N	Thigh force threshold, Han et al. (2018)	[C]	Medium	Also from the review's comparison table. Again, other studies in the same table report lower figures for the same location.
about 220 ms	Measured robot position-reporting delay	[D]	Medium	NIST measured roughly 220 ms between the robot being at a position and reporting it. Separation-based safety depends on knowing where the robot is, and that knowledge lags.

Why the numbers disagree

The largest source of disagreement is the body. Pain thresholds vary about threefold across body locations in a single study, from 196.1 N/cm2 at the back of the hand to 65.1 N/cm2 at an arm nerve. They also vary between people: the Behrens work found male and female thresholds differ at specific locations, and the reported standard deviations are wide enough to overlap several sites.

The second is method. These studies differ in how force is applied, through blunt or semi-sharp surfaces, in impact or in pinching, at different impactor shapes and speeds. Two teams measuring the deltoid reported values differing by more than double, and the review that collects them says plainly that the contact conditions and premises of the studies differ. That is a measurement problem, not a contradiction to be resolved by picking one.

The third is that force and pressure are different quantities and get quoted interchangeably. Newtons describe how hard the contact is; newtons per square centimetre describe how concentrated it is. A small, sharp contact area can be injurious at a force that would be harmless spread across a palm, which is why studies reporting N and N/cm2 cannot simply be compared.

There is also a timing dimension that force numbers hide entirely. NIST measured about 220 ms between a robot being at a position and reporting that position. Any safety approach that keeps a robot away from a person rather than limiting contact depends on knowing where the robot is now, and that knowledge is late.

How to cite these figures

Treat these as research measurements of where pain begins, not as safe operating limits. They describe human subjects in laboratory conditions, not a machine configuration.

Always name the body location and the study. A threshold without a location is meaningless when the spread across locations is threefold.

Keep force and pressure separate. Cite N and N/cm2 as different quantities, and do not convert between them without the contact area, which these studies control and most citations do not.

For anything you intend to build or commission, the normative requirements for collaborative operation live in the robot-safety standards, principally ISO 10218-1 and ISO 10218-2 with the collaborative material formerly in ISO/TS 15066. Implementing power and force limiting requires those documents and a risk assessment for your specific application. This page is not a substitute for either.

Where people go wrong

Quoting a single force figure as the threshold for human safety. The measured values differ by body location, by person, by contact geometry, and by study.

Treating a research pain threshold as a machine setting. Pain onset in a laboratory is not the same question as what a risk assessment permits for a specific cell.

Comparing a newton figure with a newton-per-square-centimetre figure. They measure different things and are not interchangeable without the contact area.

Assuming a collaborative robot is safe because it is a collaborative robot. Collaborative operation is a property of the application, set by risk assessment, not of the machine.

Reading these figures as the standard's limits. They are independent measurements, and they do not agree with each other, which is part of why the normative limits are set in a standard rather than derived from any one study.

How we checked

Every figure here comes from a peer-reviewed study accessible through PubMed Central. We retrieved each one and confirmed the figure appears in its text before publishing.

We deliberately publish only what independent research measured. The normative limits for collaborative operation are defined in robot-safety standards that we do not own and do not restate, in line with this site's practice on collaborative-safety topics: explain the concept, cite the open research, and point to where the requirements actually live.

We chose studies that report their own scale, so a reader can weigh them: 112 subjects across 28 body locations in one, 15 sites measured three times each in another. Where the review collects figures from two studies for the same body location, we report both rather than the one that suits a narrative.

We looked for a consensus dataset of human injury thresholds under robot contact and there is not one. What exists is a small number of laboratory studies with differing methods and overlapping uncertainty, and that absence is the honest headline of this topic.

Full source list

Primary sources, with live links. Every figure above traces to one of these.

[A]PMC (PubMed Central)2022
Behrens et al., biomechanical limits study, 112 subjects across 28 body locations (Frontiers in Robotics and AI), PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC8850785/
[B]PMC (PubMed Central)2019
Melia et al., "Assessment of pressure pain thresholds in collisions with collaborative robots", PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC6497239/
[C]PMC (PubMed Central)2024
Review of collision-safety measurements comparing Han et al. (2018) and Behrens et al. (2022) biomechanical results, PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC11033501/
[D]PMC (PubMed Central)2016
NIST, measurement study of speed and separation monitoring for collaborative robot systems, PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC5117641/

Common questions

How much force can a robot safely apply to a person?: There is no single answer, and this page does not provide one. Research measures where pain begins, and those thresholds vary about threefold by body location, differ between people, and differ between studies by more than double for the same location. Permissible limits for a real application are set by the robot-safety standards and a risk assessment, not by a research figure.
Why do the studies disagree?: They apply force differently, through blunt or semi-sharp surfaces, in impact or pinching, at different impactor shapes and speeds, and they test different populations. A review comparing two of them notes the contact conditions and premises differ, so the results are not directly comparable.
Which body location is most sensitive?: In the 15-site study, the arm nerve had the lowest pressure pain threshold at 65.1 N/cm2 and the back of the hand the highest at 196.1 N/cm2, roughly three times higher. Where contact happens matters more than a single force figure suggests.
Is force the only thing that matters?: No. NIST measured about 220 ms between a robot being at a position and reporting it, which matters for any approach that keeps the robot away from people rather than limiting contact. Timing and sensing lag sit alongside force in real systems.
Where are the actual limits defined?: In the robot-safety standards, principally ISO 10218-1 and ISO 10218-2, with the collaborative material formerly published as ISO/TS 15066. We do not restate their values. Implementing power and force limiting requires the standards themselves plus a risk assessment for the specific application.

More data, traced to source

What robot fatality case files actually record
The individual robot-fatality investigations that do exist point to one failure again and again: the robot was live when it should have been locked out. What the OSHA and NIOSH-family case files record.
Caught in machinery: the US severe injury and amputation numbers
Body parts caught in running machinery is the leading cause of severe workplace injury in the US, at 12,930 reports and 18% of the total. Manufacturing accounts for 69% of them. The OSHA figures, assembled.
The cobot study where output fell and posture improved
In a controlled study, workers made fewer products with a collaborative robot than without one, 5.35 versus 8.03. What improved was their posture. A measured counterexample to the cobot productivity pitch.