Loading...
E.G.

KNX Covers Without Position Feedback in Home Assistant

November 18, 2026
Part 5 of 5Series: Moving a house to a new Home Assistant instanceAll parts
  1. 01Home Assistant Migration: Old and New in Parallel
  2. 02Home Assistant Packages Instead of Include Files
  3. 03Home Assistant Backups on S3-Compatible Storage, Not AWS
  4. 04KNX Heating in Home Assistant: 21 Automations Become 6
  5. 05KNX Covers Without Position Feedback in Home Assistant
Table of Contents

The house has 43 KNX roller shutters, and on the morning of 26 September Home Assistant showed 33 of them as unknown. The other ten had an estimated position. I noticed it during the first full state comparison of the migration, between the old instance and the new one I am building next to it: all 43 covers had the same state on both, and for 33 of them that shared state was unknown. My note that morning said the position feedback was simply not linked in the ETS project and that fixing it needed a current ETS export and an ETS session. That was only partly right. Checking it first saved me a session that would have changed nothing for most of the covers.

What Home Assistant does with the covers

I had expected the covers to be a large area of the migration. Home Assistant has almost no logic of its own for them. The shutters are driven by KNX push-buttons and central functions directly on the bus, and Home Assistant only adds two time automations for one floor, which had already moved with the time-control package. So there was nothing to migrate for the covers, and the open question was what the dashboard shows for them.

Without feedback from the actuator, Home Assistant estimates a cover's position from the move command and the travel time. The estimate therefore depends on which move commands Home Assistant actually sees on the bus.

Reading the application programs from the ETS export

An ETS export, the .knxproj file, is a zip archive. It contains the group addresses and their links, and it also contains the manufacturer's complete application program for every device used in the project, as XML. That program lists all communication objects with name, size and flags, including the ones that are not active in the project. The manufacturer's ETS plugins are not needed to read it. So the export alone told me whether an actuator's program has an object for position feedback.

Hardware.xml assigns each device its program, through a Hardware2Program element that holds an ApplicationProgramRef. The program file contains the ComObject elements, and the German object names come from its Languages block. The flags are the part that matters here. An object that only has C and W, communicate and write, can only receive. To report anything back, an object needs T for transmit or R for readable.

The finding: 39 covers on actuators without a feedback object

The 43 covers are spread over three kinds of actuator. 38 of them hang on 4-channel shutter actuators, an older model from one of the large manufacturers. Its objects are up/down, stop, move to position, safety, sun protection and central, and every one of them carries only C and W. One cover hangs on a combined switch and blind actuator from the same manufacturer, with up/down, stop and lock. The remaining four are on blind actuators from a second manufacturer, and those do have an object called status display position in percent, which is just not linked in the project.

For 39 covers I found no feedback object that could be linked, so with these application programs the actuator cannot report a position. The old actuator accepts a move-to-position command and has no object to report its position. For the four covers on the other actuators the feedback is possible through ETS, and afterwards Home Assistant needs one position_state_address per cover. I have not done that part yet.

Listening to the central group addresses

Until then Home Assistant only saw each cover's own group address. In this house the shutters are often moved through central commands: all of them, one floor, one group of rooms, plus the time control in the evening. Those commands go out on separate group addresses that are linked to the actuator channel in addition to the cover's own address. Home Assistant missed those commands.

The KNX integration accepts a list in each address field. It sends on the first address and only listens on all further ones. That is enough to follow the central commands without sending anything new:

# knx.yaml — one cover, addresses are placeholders
cover:
  - name: "Cover 1"
    # first address in each list: Home Assistant sends on it
    # further addresses: central commands, listened to only
    move_long_address: ["9/2/3", "9/0/1", "9/0/2"]
    move_short_address: ["9/3/3", "9/0/11", "9/0/12"]
    travelling_time_down: 45
    travelling_time_up: 45
    # no position_state_address: the actuator has no status object

A local script reads from the export which group addresses are linked to which actuator channel. The excerpt further down shows only the function that selects the addresses. The commands after it call my local script, and its ETS parser and the code that writes the configuration are not part of this post. For each cover it adds exactly the addresses that move all of that cover's channels. Some covers drive two motors at once, and for those an address that moves only one of the two motors is left out.

# shortened from tools/knx_cover_passive.py
# ga_channels: group address -> actuator channels linked to it
# channel_gas: actuator channel -> group addresses linked to it
def passive_addresses(primary, kind, ga_channels, channel_gas):
    # the channels this cover's own address moves (can be two motors)
    target = ga_channels.get(primary, {}).get(kind, set())
    if not target:
        return None
    # every other address linked to one of those channels
    candidates = set().union(*(channel_gas[c][kind] for c in target)) - {primary}
    # keep only the addresses that move ALL channels of this cover
    return sorted(a for a in candidates if target <= ga_channels[a][kind])

Two things went wrong on the way. Object ids are only unique within one application program. My first version searched across the whole manufacturer and picked the wrong program for some channels, and the fix was to resolve the id through the concrete program of each device. The second problem was that wall push-buttons also have objects named for long operation, because they send up and down. Without a filter they were counted as channels and the intersection came out empty. The script now only looks at devices whose product name contains the word Aktor, German for actuator.

The result is 82 address fields on 41 of the 43 covers, each extended by one to five passive addresses. The sending address stays the same in every field, so nothing changes on the bus, and no telegram was sent to the bus during this work. The script also has a check mode that reports differences between the configuration and the ETS export, which I expect to use after the next export.

# derive the passive addresses from the ETS export and rewrite knx.yaml
python3 tools/knx_cover_passive.py

# change nothing, only report address fields that differ from the export
python3 tools/knx_cover_passive.py --check

What the estimate still cannot know

After a restart Home Assistant restores the last calculated position, but it knows nothing about movements that happened while it was down. Operating an actuator by hand in the distribution board produces no telegram, so Home Assistant does not notice it. The actuators' safety and sun-protection objects are separate objects and are not included in the calculation. They are currently not linked in the project, so that gap has no effect today. Separately from that, the travel times in my configuration are still flat default values for almost all covers, so a position between open and closed is a rough figure.

The test the next morning

My plan was to wait for the old instance's morning central run and watch the new instance follow it. The run never came, because that automation has been switched off on the old instance for a few days. Someone in the house pressed the central up button for one floor, and I checked the states on the new instance afterwards.

On the new instance, within two minutes, the 19 covers of that floor that had been unknown all showed open. The four covers of that floor that had been estimated as closed went to opening and then to open. The covers on the other floors did not change in Home Assistant, as expected, because their address lists contain the house-wide command and not the command for this floor. That is the only test of the mapping so far, one central up command on one floor.

Changes on the dashboard

On the dashboard, 33 of 43 cover tiles had been showing the word Unknown. The tiles now carry no state text at all, which is the hide_state option of the built-in tile card. The icon colour shows the estimate, teal for probably open, grey for probably closed and pale for unknown. Home Assistant colours open covers purple by default, which did not fit the palette. A note below the tiles says that up, stop and down can be sent whatever the icon shows. I had made the same decision for the heating view, where the room controllers do not report their operating mode, which is why I removed the mode selector from the overview.

I also removed a helper sensor that counted open covers. It counted estimates, nothing used it, and at some point it could have misled an automation. Once the four covers with a status object have their feedback linked in ETS and their position_state_address set in Home Assistant, I will remove hide_state from those tiles.

Frequently asked questions

Can I tell from the ETS project alone whether an actuator supports position feedback?

Yes. In my export the application XML of each device showed whether a position status object exists and whether its flags allow sending or reading. I did not need the manufacturer's plugin.

Do passive addresses change anything on the KNX bus?

No, as long as the first address of each list stays the same. Home Assistant sends only on that one and listens on the others.

What does Home Assistant show for these covers after a restart?

The last calculated position, marked as an assumed state. If there is nothing to restore, the state is unknown until Home Assistant sees a movement command.

Related articles

Questions or additions?

Comments run on GitHub Discussions, so you need a GitHub account to write one.