There are two protocols involved in the communication between the client and Ecovacs systems. There are a series of HTTPS requests used to log in and find devices. Once logged in, you get a token that is used to connect to an XMPP server, which mediates communication with the vacuum. That's right, your robot housecleaner, like an errant teen, spends all its free time hanging out in an internet chat room.
This is all taken from MITMing the Android app. The iOS app appears to follow the same protocol conventions.
It appears that Ecovacs have broken up their API servers by location. Some are designated by country, others by continent. All appear to use the two-letter ISO codes, but at this time it doesn't look like all codes map to valid servers.
The HTTPS and XMPP servers do not appear to be following the same convention. For example, a Canadian user must authenticate on country-specific HTTPS server, but XMPP commands work both on the worldwide server msg-ww.ecouser.net) and the North America server (msg-na.ecouser.net)
The Android App uses the following XMPP messaging servers:
CH: msg.ecouser.net
TW, MY, JP, SG, TH, HK, IN, KR: msg-as.ecouser.net
US: msg-na.ecouser.net
FR, ES, UK, NO, MX, DE, PT, CH, AU, IT, NL, SE, BE, DK: msg-eu.ecouser.net
Any other country: msg-ww.ecouser.net
There are two sorts of URLs in the basic login flow. The first set use a format like this:
https://eco-{country}-api.ecovacs.com/v1/private/{country}/{lang}/{deviceId}/{appCode}/{appVersion}/{channel}/{deviceType}
They also have a complicated API request signature that seems overelaborate to me. See the Python code for more details.
- GET eco-us-api.ecovacs.com ... common/checkVersion - appears to just check the app version
- GET eco-us-api.ecovacs.com ... user/login - Sends encrypted versions of the username and password. The response is some json containing a uid and access token.
- GET eco-us-api.ecovacs.com ... user/getAuthCode - sends uid, accessToken; gets back an auth code
Now we switch to posting to a different server, and the request and response style change substantially. I think of this at the user server, or perhaps the XMPP/device server.
- POST users-na.ecouser.net:8000/user.do loginByItToken - trades the authCode from the previous call for yet another token
- POST ne-na.ecouser.net:8018/notify_engine.do - not sure what this is for; my script skips this and seems to work fine
- POST users-na.ecouser.net:8000/user.do GetDeviceList - Using the token from step 4, gets the list of devices; that's needed for talking to the vacuum via XMPP
Under mysterious circumstances, for some people the getAuthCode call will return a different userId than is passed in. In that case, apparently the new userId should be used for future calls, or an Auth 1004 error results.
The app establishes a connection to an XMPP server and logs in using a secret that comes from the earlier HTTPS calls. It then sends XMPP IQ commands. It describes them as queries, but they all contain "ctl" elements that appear to be commands.
Command
<ctl td="Clean"><clean type="auto" speed="standard"/></ctl>
State
- Request
<ctl td="GetCleanState" />
- Response
<ctl td="CleanReport"><clean type="stop" speed="standard" /></ctl>
- type
auto
automatic cleaning program - type
border
edge cleaning program - type
spot
spot cleaning program - type
singleroom
cleaning a single room - type
stop
bot at full stop - speed
standard
regular fan speed (suction) - speed
strong
high fan speed (suction)
- type
Command
<query xmlns="com:ctl"><ctl td="Charge"><charge type="go"/></ctl>
go
order bot to return to charger
State
- Request
<query xmlns="com:ctl"><ctl td="GetChargeState" />
- Response
<ctl td="ChargeState"><charge type="SlotCharging" /></ctl>
Idle
not trying to chargeGoing
trying to return to chargerSlotCharging
currently charging in dockWireCharging
currently charging by cable
Battery charge level. 080 = 80% charged. State is broadcast continously when the robot is running och charging, but can also be requested manually.
- Request
<ctl td="GetBatteryInfo" />
- Response
<ctl td="BatteryInfo"><battery power="080" /></ctl>
The remaining lifespan of components. Based on an internal counters that can be reset with command ResetLifeSpan (untested).
It's presumed that the timers need to be reset manually.
- Request
<ctl td="GetLifeSpan" type="Brush" />
- Response
<ctl td="LifeSpan" type="Brush" val="095" total="365" />
- Brush
- SideBrush
- DustCaseHeap
Command
- Move forward:
<ctl td="Move"><move action="forward"/></ctl>
- Spin left 360 degrees:
<ctl td="Move"><move action="SpinLeft"/></ctl>
- Spin right 360 degrees:
<ctl td="Move"><move action="SpinRight"/></ctl>
- Turn 180 degrees:
<ctl td="Move"><move action="TurnAround"/></ctl>
- Stop the ongoing action:
<ctl td="Move"><move action="stop"/></ctl>
Set/get robot internal clock
<ctl td="SetTime"><time t="1509622697" tz="+8"/></ctl>
<ctl td="GetTime"></ctl>
- Time is specified as a UNIX timestamp and timezone + or - UTC offset.
Get firmware version
<ctl td="GetVersion" name="FW"/>
Get robot logs
<ctl td="GetLog"></ctl>
The bot broadcasts error codes for a number of cases.
<ctl td="error" error="BatteryLow" errno="101"></ctl>
The latest error can be requested like so:
- Request
<ctl td="GetError" />
- Response
<ctl ret="ok" errs="100"/>
However in some cases the robot sends to code 100 shortly after an error has occurred, meaning that we cannot trust the GetError request to contain the last relevant error. For example, if the robot gets stuck it broadcasts 102 HostHang, then proceeds to stop and broadcasts 100 NoError.
Known error codes
- 100 NoError: Robot is operational
- 101 BatteryLow: Low battery
- 102 HostHang: Robot is stuck
- 103 WheelAbnormal: Wheels are not moving as expected
- 104 DownSensorAbnormal: Down sensor is getting abnormal values
- 110 NoDustBox: Dust Bin Not installed
These codes are taken from model M81 Pro. Error codes may differ between models.
<ctl td="SetOnOff" t="b" on="1"/>
<ctl td="GetOnOff" />
<ctl id="12351409" td="PlaySound" sid="0"/>
<ctl id="30800321" td="GetSched"/>
It appears that it adds an extra id when it cares to receive a specific response. This is a little odd in that the iq blocks already contain ids, but perhaps one is more a server id and the other is used by the robot itself.