Where a figure has not been measured, this page says so rather than estimating. A number you can plan a flight against has to come from somewhere, and a plausible number with no measurement behind it is worse than an admitted gap.
Measured: the radio link
From two separate bench runs on a paired drone and ground station, withados diag link and the receiver’s own counters as independent sources:
Read these for what they are: healthy links at close range, confirming the pipeline decodes cleanly end to end. Neither is a range figure, and the throughput in each is what that session happened to be carrying rather than a ceiling.
Two things they do establish. Zero decrypt errors alongside a recovered-FEC count means the crypto and error correction are working on real radio frames, not on a loopback. And a signal this strong with zero loss is the shape of a healthy short-range link, which is the baseline to compare a degraded one against. The second run reached the same shape independently, after both ends were re-paired from scratch.
The agent reports a link as degraded once loss passes 50 percent or signal falls below -85 dBm.
Measured: publish-side video latency
One video latency figure has been measured, and it is not glass-to-glass. A timestamp probe injected after the encoder and read back from the drone’s own RTSP stream measured a mean of 74 ms (range 61 to 152 ms, 6 samples) on a sim bench at 1280x720, 30 fps, 4000 kbps. That figure covers only the publish mux, the RTSP publish, the media server on the drone, and the readback. It excludes capture, encode, the radio hop, the ground receive, the ground media server, and the browser, which are most of what an operator experiences. There is no glass-to-glass latency figure for ADOS. Do not add per-stage guesses to the 74 ms to manufacture one. The method and its caveats are on Video stack.Not characterised
Each of the following needs controlled flight testing that has not been done. Do not plan against a number that is not here.Range
Not characterised by our own flight testing. The WFB-ng page describes what the link can do with directional antennas. No measured range test on an ADOS aircraft exists that records the distance at which the link degrades with the antennas, transmit power, and terrain alongside it. Range depends on antenna gain and pattern on both ends, transmit power, terrain and line of sight, interference, and the regulatory limit where you fly. A single number without those conditions attached would not transfer to your setup. If you need a range figure you can rely on, fly it yourself, incrementally, with the aircraft in sight and a working RC failsafe.Endurance
Not characterised. Flight time is a property of your airframe, battery, and payload, not of ADOS. The companion computer and radio draw power that shortens it, and that draw has not been characterised per board.Altitude
No altitude ceiling has been measured, and ADOS does not impose one. Your altitude limit is whatever your jurisdiction sets and whatever you configure inFENCE_ALT_MAX. See Regulatory.
Link budget
Not characterised. No measured curve of signal strength against distance, transmit power, and antenna gain exists for ADOS hardware, so there is no link budget table to publish. The -36 dBm reading above is one point at close range, not a curve.Client capacity
Roughly 3 to 4 devices can stream video at once over the ground station’s 2.4 GHz access point before throughput degrades. This is a working figure from bench use, not a characterised limit. Telemetry over WebSocket is far lighter and supports more clients.Limits ADOS enforces
These are in the code, unlike the figures above.Video defaults
The full-rate (hero) profile defaults to 1280x720 at 30 fps, 4000 kbps (video.camera). The thumbnail profile defaults to 320x180 at 1 fps, 50 kbps (video.camera.thumbnail), so that adding aircraft costs a small fraction of the channel rather than another full video stream. A drone on the WFB radio boots on its thumbnail profile, so a fleet powering up together does not flood the channel. It runs hero while anyone holds a lease on it: Mission Control takes and renews a lease (5 to 120 seconds, 30 by default) while a detail view such as the cockpit plays the drone’s feed, and the drone drops back to thumbnail once the last lease is released or expires. Fleet tiles never take a lease. While a vision engine is attached, the agent holds its own vision lease, so the vision tap never reads a thumbnail-rate stream.
Fleet size
The radio addressing scheme supports up to 24 aircraft on one ground station.Guided setpoints decay
Anything commanding the aircraft through guided setpoints, including follow behaviours and swarm control, has to keep sending. Most autopilots brake roughly 3 seconds after the last setpoint and then run their own failsafe. A single setpoint does not hold; it decays. This is a property of the flight controller, not of ADOS, and it is the reason a companion-side behaviour that stops emitting causes the aircraft to hold position rather than continue.Transmit power and brownout
The transmit power clamp is a hardware protection, not a regulatory one, and it stays in force in every operating region posture. Raising transmit power past what the adapter’s power supply can carry does not increase range, it drops the link. If you run an adapter on a powered hub or external 5 V rather than host power, setvideo.wfb.topology accordingly so the brownout warning reflects your setup.
The exact power at which a given adapter browns out has not been characterised. It depends on the adapter, the cable, and the supply, so treat the ceiling as the tested configuration rather than a measured cliff.
Next steps
Regulatory
The limits that are legal rather than physical.
Hardening
The credentials a node holds and how to lock it down.
Video stack
How the latency figure was measured and what it covers.
WFB-ng
The radio link, its configuration, and its counters.