Free tool

Teleoperation latency calculator

Estimate the round trip a remote operator will feel, from camera to screen and command back to the robot, and see which control mode each robot type supports at that delay. It's an estimate: measure glass-to-glass on your real link before you commit.

  • Runs in your browser. Nothing is sent anywhere
  • Updated
  • 24 cited sources

Your link

Distance from
Network at the robot
Adds this much round trip · typical, see [3]ms
Network at the operator
Adds this much round trip · typical, see [3]ms
ms

Capture, encode, decode, display and the robot's controller. 60 ms assumes a tuned low-latency stream: measure yours.

×

Fiber rarely follows the great circle; 1.5 is a common planning value.

Command VideoShown 10× slower · drag to turn
5,570 km great circle · light alone 55.7 ms there and back

Estimated round trip

≈164ms

Band 150–300 ms

Breakdown, ms round trip
Light in fiber, with route factor83.651%
Network at the robot106%
Network at the operator106%
Video and control pipeline6037%

What ≈ 164 ms supports

What the modes mean

Rules of thumb from published studies, not hard limits: operator skill, the interface, jitter and on-robot assistance all shift them. A steady delay is easier to work with than one that swings.

Book a RoboPilot

Humanoid

At ≈ 164 msDirect control

No commitment. Your quote arrives within one business day.

Where the milliseconds go

A round trip has four parts. Light in fiber is the floor nobody negotiates: about 200 km per millisecond, so every 1,000 km of route adds about 10 ms there and back. The two access networks add the last hop at each end and most of the jitter. The video and control pipeline (camera, encoder, decoder, display and the robot's controller) is often the largest single term on short links, and the one you control most.

RobotCapture, encode, controller30ms
Robot networkLast hop at the robot10ms
Long-haul fiber5,570 km × 1.583.6ms
Operator networkLast hop at the desk10ms
OperatorDecode, display, react30ms
Video travels 1 → 5, the command comes back 5 → 1. The pipeline's 60 ms is split across both ends.≈ 164 ms round trip

Feel the delay

Numbers undersell latency. Move your pointer or finger in the pad: the square is the robot as the operator sees it, one round trip behind. Try 40 ms, then 800 ms, and notice the moment you stop steering and start waiting.

Your hand The robot, as you see it
Move here. The robot follows 164 ms behindArms at this delay: direct control

The four control modes

As the delay grows, the way an operator works with the robot changes. The calculator places each robot type in one of four modes, using the studies cited in the teleoperation latency guide.

Direct control4 of 6 robot types now

Continuous, moment-to-moment control works. The operator moves the robot by joints, hand pose or sticks and sees each result in time to correct it.

Assisted control2 of 6 robot types now

Continuous control still works, but slower and less precise. The robot's own controllers and safety limits carry more of the fine motion.

Move and wait

Continuous control becomes hard. Operators move in short steps and wait to see each result: slower, but reliable.

Supervision

The robot acts on its own. The operator sets goals, waypoints and approvals and handles exceptions, so delays of seconds are workable.

What the operator commandsWhat the robot does, over time

Every band, every robot type

The rules behind the verdicts: six round-trip bands, six robot types, and the sources for each cell. Read across a row to see where a robot type changes mode; your link's band is highlighted.

Control mode supported by each robot type, per round-trip band
Round trip< 50 ms50–150 ms150–300 msYour link300–700 ms700–1000 ms> 1000 ms
Robot armDirect control[14]Direct control[10] [9]Direct control[11] [13]Assisted control[11] [12]Move and wait[11] [12]Move and wait[7] [8]
HumanoidDirect control[15]Direct control[16] [15]Direct control[15] [17]Assisted control[11] [12]*Move and wait[11] [12]*Move and wait[7] [8]*
QuadrupedDirect control[7]*Direct control[18]Assisted control[7] [19]*Supervision[6] [21]*Supervision[6] [21]*Supervision[21]*
Wheeled robot or vehicleDirect control[7]Direct control[7] [19]Assisted control[7] [20]Supervision[6] [21]Supervision[6] [21]Supervision[21]
DroneDirect control[22]Direct control[22] [23]Direct control[22] [23]Assisted control[23] [24]Supervision[22] [23]Supervision[22]
Other robotDirect control[14]*Direct control[10] [9]*Direct control[11] [13]*Assisted control[11] [12]*Move and wait[11] [12]*Move and wait[7] [8]*
* Evidence from a similar robot type. A band's lower edge is inclusive: 150 ms falls in 150–300.

Measure it for real

Point a robot camera at a running millisecond clock shown on the operator's screen, then photograph both together. The difference is your video latency. Track percentiles and jitter, not only the mean, and measure where the robot actually works.

1 · Clock on the operator's screen

00:07.000

2 · Robot camera's view of it

00:06.900

Photograph both at once: the difference is your one-way video delay. Simulated here at 82 ms, half your estimated round trip.

Method

Round trip =

propagation + robot-side network + operator-side network + video and control pipeline

  • DistanceGreat-circle (haversine) distance between the two cities, Earth radius 6,371 km.
  • PropagationLight in optical fiber travels at about 2/3 of its speed in a vacuum, roughly 200 km per millisecond, or 5 µs per km (ITU-T G.114 uses 5 µs/km). Round-trip propagation = 2 × distance × route factor ÷ 199.9 km/ms. Real fiber paths are longer than the great circle, so distance is multiplied by a route factor, 1.5 by default.
  • Access networksThe last hop at each end adds delay and most of the jitter. The defaults below are typical medians; loaded or congested links can be several times higher.
  • PipelineCamera exposure, encoding, decoding, display refresh and the robot's controller. Often the largest single term on short links, and the one you control most.
  • Measured pingA ping between the two sites replaces propagation and both access networks.
  • Bands and modesSix round-trip bands (< 50, 50–150, 150–300, 300–700, 700–1000 and > 1000 ms), each with the control mode the evidence supports per robot type. A band’s lower edge is inclusive.

Network defaults

NetworkRound tripBasis
Fiber / cable (wired)10 msWired broadband adds the least. The FCC Measuring Broadband America report tracks idle and loaded latency by ISP and technology, with fiber lowest. 10 ms is a planning value for a wired line to a nearby server. [3]
5G35 msOokla 2026: median 5G multi-server latency ranged from 24.6 ms (Singapore) to about 53 ms across 22 markets; loaded latency can reach hundreds of ms. [4]
4G LTE60 msRemote-driving trials over LTE report round-trip minimums around 45 ms, rarely under 50 ms, and often 80 to 100 ms (Kamtam et al. 2024 review). [6]
Starlink (LEO satellite)40 msOokla: Starlink US median multi-server latency 39 ms in Q1 2026, with some states around 37 ms. [5]
GEO satellite650 msOokla Q1 2026 medians: Hughesnet 674 ms, Viasat 750 ms. Physics floor: a geostationary hop is about 36,000 km each way, about 480 ms round trip before any processing (ITU-T G.114 lists 260 ms one-way for a 36,000 km satellite hop). [5, 2]
Already counted / LAN0 msUse when the operator sits on the same local network as the robot, or you have counted this end elsewhere.

Questions

What is a good latency for teleoperation?
It depends on the robot. In the studies behind this calculator, robot arms, humanoids and drones keep direct, moment-to-moment control up to about 300 ms round trip. Wheeled robots and quadrupeds move to assisted control from about 150 ms and to supervision from about 300 ms.
How much latency does distance add?
Light in optical fiber covers about 200 km per millisecond, so round-trip propagation is 2 × distance × route factor ÷ 199.9 km/ms. New York to London (about 5,570 km) gives about 56 ms of pure fiber round trip, or about 84 ms with the default 1.5 route factor.
Is 5G or Starlink fast enough to teleoperate a robot?
By median latency, both add tens of milliseconds: the defaults here are 35 ms for 5G and 40 ms for Starlink, from Ookla's 2026 figures. Loaded latency can be several times higher, and a steady link matters as much as a fast one. A geostationary satellite adds about 650 ms, which moves most robot types to move-and-wait or supervision.
Can I measure my link instead of estimating it?
Yes. Enter a measured ping between the two sites and it replaces distance and both access networks. For the video path, point a robot camera at a millisecond clock on the operator's screen and photograph both: the difference is your glass-to-glass delay.
Does the calculator send my inputs anywhere?
No. It runs entirely in your browser. Your inputs are kept only in the page address, so you can share a result as a link.

Sources

  1. [1]RoboPilots, Teleoperation latency: how much delay each robot task tolerates (2026)
  2. [2]ITU-T Recommendation G.114, One-way transmission time
  3. [3]FCC, Measuring Broadband America: Thirteenth Fixed Broadband Report
  4. [4]Ookla, Benchmarking 5G for AI workloads (2026)
  5. [6]Kamtam et al., Network Latency in Teleoperation of Connected and Autonomous Vehicles: A Review, Sensors 2024
  6. [7]Chen, Haas, Pillalamarri & Jacobson, Human-Robot Interface, ARL-TR-3834 (2006)
  7. [8]Ferrell, Remote manipulation with transmission delay (summarized in Black, MIT 1970)
  8. [9]MacKenzie & Ware, Lag as a determinant of human performance in interactive systems, INTERCHI 1993
  9. [10]Anvari, McKinley & Stein, Establishment of the world's first telerobotic remote surgical service, Ann Surg 2005
  10. [11]Xu et al., Latency effects on surgical performance and acceptable latency levels in telesurgery, Surg Endosc 2014
  11. [12]Perez et al., Impact of delay on telesurgical performance, Int J CARS 2016
  12. [13]Fong et al., Clinical Recommendations for Remote Robotic Assisted Surgery (CRSA 2025), World J Surg 2026
  13. [14]Xu, Liu & Steinbach, QoE-driven control scheme switching for time-delayed teleoperation, 2017
  14. [15]Xiong et al., ExtremControl: Low-Latency Humanoid Teleoperation with Direct Extremity Control, 2026
  15. [16]Xu et al., Beyond Retargeting: Low-Latency and Robust Humanoid Whole-Body Teleoperation, 2026
  16. [17]Schwarz, Lenz et al., NimbRo wins ANA Avatar XPRIZE Finals, IEEE-RAS Humanoids 2023
  17. [18]Risiglione et al., Passivity-based control for haptic teleoperation of a legged manipulator with time delays, IROS 2021
  18. [19]Chen et al., Effects of communication delay on human performance in mobile robot teleoperation, 2025
  19. [20]Neumeier et al., Teleoperation: The Holy Grail to Solve Problems of Automated Driving? Sure, but Latency Matters, AutomotiveUI 2019
  20. [21]SAE J3016 remote assistance vs remote driving (as quoted in Bogdoll et al. 2021)
  21. [22]3GPP TS 22.125 V17.6.0, UAS command and control latency (one-way figures, read conservatively as round trips)
  22. [23]Zingale & Taylor, Effect of Control Latency on Unmanned Aircraft Systems During Critical Phases of Flight, ISAP 2015
  23. [24]Cox & Wong, Predictive feedback augmentation for manual control of a UAV with latency, IJMAV 2019
Need operators who work well at your latency? Tell us about your robots.