Industrial Wi-Fi for AGVs and AMRs: Roaming and Latency Design

Industrial Wi-Fi for AGVs and AMRs: roaming time, 802.11r/k/v, channel planning among shelving, SSID and QoS separation, and acceptance testing while moving along the route.
Technical summary
Technology
IEEE 802.11r, 802.11k, 802.11v, WMM, WPA2/WPA3, 5 GHz, DFS
Topic
Wireless roaming and latency design for mobile robots
Level
Intermediate–advanced
Prerequisites
  • Wi-Fi channel and RSSI concepts
  • Basic knowledge of VLANs and QoS
  • The AGV/AMR manufacturer's network requirements document
In this article8 sections

The robot stops in the middle of an aisle, the screen says “communication timeout”, the operator resets it and the robot continues on its way. The next day, the same fault at the same spot. On office Wi-Fi, nobody notices a dropout of a few hundred milliseconds; for AGVs and AMRs, the same dropout means an emergency stop, a lost route or a cancelled task. Below you will find the order of checks that works in the field for roaming, latency budget, RF environment and acceptance testing when designing Wi-Fi for a robot fleet in warehouse and production areas.

What robot traffic expects from the network

Robot traffic is low-bandwidth but time-sensitive. Periodic position and status messages to the fleet manager, task assignments, traffic management (intersection and lift locks), map updates, remote maintenance sessions and, on some models, camera video all flow over the same link. Among these, the single value that drives the design is the communication timeout defined by the manufacturer: if the robot receives no response from the fleet manager for a certain time, it slows down, stops or switches to safe mode.

Get this value from the manufacturer in writing. A verbal answer such as “it tolerates a few seconds” is not a design input. The Wi-Fi connection carries the robot’s task and status data; safety functions such as collision avoidance and emergency stop remain in the robot’s own safety system and do not rely on the wireless connection. Information to request:

  • Control message period and timeout (ms)
  • Average and peak bandwidth per robot
  • Protocols and ports used (TCP/UDP, whether multicast is used)
  • The bands, channels, security methods and 802.11r/k/v support of the Wi-Fi module
  • Whether the robot’s roaming threshold is configurable

A design made without this list does not verify the robot’s behaviour on its actual route, even if the heat map looks adequate.

Roaming time and the client’s decision

In Wi-Fi, the roaming decision is made by the client, not the access point. The robot’s Wi-Fi module sees that the current AP’s signal has fallen below the threshold, scans other channels, selects a candidate AP, completes authentication and key exchange, and then resumes sending data. The total interruption is the sum of these steps; the largest items are usually scanning and authentication.

A common problem is the “sticky client”. Because the threshold is set too low, the robot clings to a weak, distant AP instead of the strong AP it is passing; the data rate drops, retransmissions increase, and when the connection drops, roaming happens in an emergency. The opposite also causes problems: if the threshold is too high, the robot keeps bouncing between two APs (ping-pong). In both cases, the AP usually gets the blame. The real problem is the client setting.

If the roaming threshold is configurable, it should be chosen to match the cell-edge signal level in the coverage design. An example approach: target at least -67 dBm for the primary AP and at least -75 dBm for the secondary (roaming candidate) AP along the robot’s route, and set the robot’s roaming threshold between these two values. These values are examples; the final value is determined by the sensitivity of the robot’s module and the manufacturer’s recommendation.

What 802.11r, 802.11k and 802.11v do

The three amendments speed up different steps of the roaming process. One does not replace another:

Standard Step affected What it means for AGVs
802.11r (Fast BSS Transition) Authentication and key derivation Eliminates full EAP re-authentication with 802.1X
802.11k (Radio Resource Measurement) Scanning With a neighbour AP list, the client does not have to scan every channel
802.11v (BSS Transition Management) Decision The AP can suggest a more suitable AP to the client; the decision still rests with the client

On a robot SSID using 802.1X/EAP without 802.11r, a full authentication involving a round trip to the RADIUS server takes place at every roam. This time varies with network and server load and easily exceeds the control message timeout. On WPA2-Personal (PSK) networks, only the four-way key exchange takes place during roaming, so the gain from FT is smaller; on WPA3-Personal (SAE), if the PMK cache cannot be used, an SAE handshake is added to every transition and FT’s effect grows. The choice of security method directly changes roaming time.

The RF environment in the warehouse and the site survey

Metal shelving, stock, liquid-filled packaging and moving forklifts change the RF environment throughout the day. A measurement taken in an empty warehouse loses its validity when the shelves fill up. Points to check during the survey:

  • Stock levels: If possible, measure during both low and high occupancy periods; if not, measure the sections that represent the worst case separately.
  • Antenna height: Place the measurement client at the real height of the robot’s antenna. A laptop carried at shoulder height does not see what the robot sees.
  • Aisle layout: In narrow, long shelf aisles, placing APs at the ends of aisles with directional antennas can give more predictable coverage than omnidirectional APs in the middle of the ceiling.
  • Channel plan: For the robot SSID, a 20 MHz channel width at 5 GHz provides more non-overlapping channels and lower co-channel interference. Robot traffic does not need wide channels.
  • DFS channels: When radar is detected, the AP must leave the channel; moving to a new DFS channel requires a listening period. On a robot’s route, this means a sudden dropout. If possible, plan the robot network on non-DFS channels.

The survey output should not be just a heat map. A report is required showing, at every point on the robot’s route, the primary and secondary AP signal, co-channel interference and signal-to-noise ratio. If there is no secondary AP value, there is no roaming design either. For details of roaming mechanisms on the enterprise network side, the article 802.11k, 802.11v and 802.11r roaming mechanisms is a useful complement.

SSID, VLAN and QoS separation

Every additional SSID means a separate beacon broadcast on every AP and consumes airtime. Even so, a separate SSID for robots is usually the right choice: the security method, FT setting, minimum data rate and QoS profile are set separately according to robot requirements. The structure we recommend:

  • Robot SSID: 5 GHz only, a separate VLAN, access rules allowing only the necessary ports between the robots and the fleet manager.
  • Handheld terminal and tablet SSID: Separate VLAN; not placed in the same priority class as robot traffic.
  • Guests and personal devices: Not broadcast in the production area, or bandwidth-limited.

On the QoS side, WMM’s four access categories (AC_VO, AC_VI, AC_BE, AC_BK) are used. Map robot control traffic to a higher-priority category than camera traffic and file transfers. Putting all robot traffic into AC_VO, however, also prioritises large transfers such as map downloads; the control message still waits in the queue. If the robot does not do its own DSCP marking, marking can be done on the wired side based on the fleet manager ports. For the robot VLAN’s address plan, the cell-based approach in the article IP addressing and naming standards for factory networks can be used.

In industrial network infrastructure, roaming should stay at Layer 2. The robot’s IP address should not change when it moves between APs connected to different subnets; the robot VLAN should be the same along the entire route, or in a controller-based architecture, traffic should be tunnelled to a central point. Otherwise, the client renews DHCP. The interruption goes up to seconds.

Measurement on the move and acceptance testing

Acceptance testing is not done with a static speed test; it is done on the robot’s real route and at its real speed. Timestamped measurements can be taken with a Linux client fixed to the robot as follows:

# Connected AP (BSSID), signal and data rate
iw dev wlan0 link

# Timestamped ping at 20 ms intervals (short intervals may require root)
sudo ping -D -i 0.02 -W 1 10.20.30.10 | tee route-test.log

# Record association/disassociation (roaming) events with timestamps
sudo iw event -t | tee roaming-events.log

Consecutive losses and latency spikes in the ping log are matched by timestamp with the iw event output and the roaming event logs of the AP or controller. For this, the clocks of the test client and the controller must be synchronised with NTP; if the clocks drift, matching turns into guesswork. The interruption at each roaming event must stay below the robot’s timeout value with a safe margin. Example calculation: if the manufacturer’s timeout is 500 ms and the worst roaming interruption measured is 150 ms, the margin is sufficient; if it is 400 ms, a single retransmission or load increase can stop the robot.

Acceptance test checklist:

  • All route segments tested at least twice, in both directions and at real speed
  • The interruption time was recorded at every AP transition; the worst value was reported
  • Primary and secondary AP signal values along the route met the design target
  • One AP was switched off and coverage and roaming with the neighbouring APs were measured again
  • The test was repeated during peak shift activity (with forklift and handheld terminal traffic present)
  • DFS events, channel changes and AP restart logs were reviewed for the duration of the test

Order of checks for connection drops

When a robot stops, the first thing to look at is whether the event is tied to a location or to a time. If it recurs at the same spot, it is RF and roaming; if it affects the whole fleet at random times, it is channel and controller events. The table follows this distinction:

Symptom Likely cause Check
Stops recurring at a specific location Coverage gap or late roaming Secondary AP signal and roaming threshold at that location
All robots affected at random times DFS channel change or automatic RF optimisation Controller channel change logs
Dropouts lasting seconds after roaming Different VLAN or subnet, DHCP renewal VLAN consistency along the route
Problem only with certain robot models FT or security setting incompatibility, driver version Client connection logs and driver version
Dropouts increasing when shelves are full Stock effect not taken into account in the survey Re-measurement when full

Automatic channel and power adjustment (RRM) changing channels during production hours is also a common cause. Fixing the channel and power on APs in the robot area, or limiting automatic adjustment to the maintenance window, makes dropouts predictable. For the robustness and PoE budget of access switches on the wired side, see the industrial switch selection guide, and for separating the network from IT, the article IT/OT network separation with the Purdue model.

For wireless design, site surveys and acceptance testing on the move for a robot fleet, our industrial wireless and mobile terminals service can help, and on the wired backbone side, our industrial Ethernet and factory networks work. The current list of standards is on the IEEE 802.11 working group page.

Frequently asked questions

Can the 2.4 GHz band be used for AGVs?

Technically yes, but we do not recommend it for control traffic. At 20 MHz there are in practice only three non-overlapping channels (1, 6, 11), and interference from Bluetooth and other devices is high. If the robot client supports it, choose 5 GHz and non-DFS channels; keep 2.4 GHz separate for legacy devices that require it.

Does enabling 802.11r always shorten roaming time?

Only if the client implements 802.11r correctly. On networks using 802.1X, the gain is large; on PSK networks, the difference is small. Drivers that do not support it or implement it incorrectly may not be able to connect to an FT-enabled SSID at all. First run a pilot test with the robot’s own Wi-Fi module.

Which method is most reliable for measuring roaming time?

Collect a loss and latency series with frequent, timestamped pings from the robot’s own client or an equivalent client fixed to the robot; match it with the roaming event logs of the APs or controller. In a wireless packet capture, the time between the last data frame on the old AP and the first on the new AP can also be measured directly.

Is it a problem to share the AGV SSID with the existing handheld terminal SSID?

A shared SSID forces the same security method, roaming settings and QoS profile on both groups of devices. Because robots have a tighter latency requirement, a separate SSID and a separate VLAN give a more manageable setup in terms of both configuration flexibility and fault diagnosis.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Industrial

PLC, PROFINET/PROFIBUS and SCADA communication, IT/OT network segmentation and industrial field connectivity. We install and monitor without stopping the production line.

Knowledge Center

Other articles