Answers checklist.
General issue report
This might not be the right place for this, but I'm not sure where else to post it.
I've been testing out the fairly new esp_wifi_sensing component and while the setup and usage were fairly pain-free, I have some issues with the training and presence detection.
I've been training it for about a minute after boot in an empty room with closed doors, and the wander threshold it produces is usually so low that it detects a presence immediately after training ends. I've played around with the presence_sensitivity parameter in the channel config, but it seems to only make the threshold smaller, not larger. For example, when calibration returns a wander threshold of 0.0014, a very low presence_sensitivity of 0.01 gives a detection threshold of approximately 0.0011. At the maximum value (1.0) the threshold drops to approximately 0.0000.
Not sure if this is a problem with the calibration itself, maybe the recommended calibration period should be longer?
I've kind of worked around the issue by implementing my own presence decision: instead of using presence_someone_status from the diagnostic struct, I double the training threshold and compare that against presence_wander_average. That seems to work relatively well in my environment, but I'm not sure if it's just a sheer coincidence or if this is the correct way to solve the issue.
All tests where done on ESP32-C6 and esp-idf version 5.5.1.
Answers checklist.
General issue report
This might not be the right place for this, but I'm not sure where else to post it.
I've been testing out the fairly new esp_wifi_sensing component and while the setup and usage were fairly pain-free, I have some issues with the training and presence detection.
I've been training it for about a minute after boot in an empty room with closed doors, and the wander threshold it produces is usually so low that it detects a presence immediately after training ends. I've played around with the presence_sensitivity parameter in the channel config, but it seems to only make the threshold smaller, not larger. For example, when calibration returns a wander threshold of 0.0014, a very low presence_sensitivity of 0.01 gives a detection threshold of approximately 0.0011. At the maximum value (1.0) the threshold drops to approximately 0.0000.
Not sure if this is a problem with the calibration itself, maybe the recommended calibration period should be longer?
I've kind of worked around the issue by implementing my own presence decision: instead of using presence_someone_status from the diagnostic struct, I double the training threshold and compare that against presence_wander_average. That seems to work relatively well in my environment, but I'm not sure if it's just a sheer coincidence or if this is the correct way to solve the issue.
All tests where done on ESP32-C6 and esp-idf version 5.5.1.