PCA9685 with ESP32 and wifi performance - Watchdog timeout

[sjgcornish]

servoEasing is a very nice and clever bit of software however when using PCA9685 and controlling the servos over WIFI there is an unpredictable lag between command and action. I am not entirely sure how servoEasing works using interrupts for servo sync, my thoughts are that he PCA has its own 50Hz clock and it sends its stored commands to the servos on a 20ms loop, this means that you only have to send a command via I2C when needed to make a change in position such as when receiving a command over WIFI. I assume that servoEasing has its own 20ms interrupt which syncs the servos and it is also updating all the the servos every 20ms, which is required if you are not using the PCA9685. For some reason it appears that, WIFI, servoEasing with PCA9685 functionality do not play well together, perhaps it is related to yield() not working in the 'while' loop. One solution for me is to use a dedicated non WIFI Arduino for servo control and just send it commands from the ESP32 over serial but then you need call back to the ESP32 when the move is completed (I have implemented this mechanism using the ESPservo library)

Or could it be due to ..
// The ESP32 I2C interferes with the Ticker / Timer library used.
// Even with 100 kHz clock we have some dropouts / NAK's because of sending address again instead of first data.

[ArminJo]

Please let a Led change its state if you received a command from wifi and issuing a command for servo easing to determine if wifi lags.
Do you use blocking wait functions for servo movements? Maybe enabling line 1616 and 1628 in ServoEasing.hpp helps in this case.
Please report.

[ArminJo]

I have no clue, but you can try inserting a yield after the line 3413 in ServoEasing.hpp

for (uint_fast8_t tServoIndex = 0; tServoIndex <= ServoEasing::sServoArrayMaxIndex; ++tServoIndex) {
yield();

[ArminJo]

Typo! It is 2413. Function updateAllServos()

[sjgcornish]

Getting closer to the problem at least for the unplanned resets. I am using webserial.h to send move commands and the error that chrome dev tools reports is ..

'Unchecked runtime.lastError: The message port closed before a response was received'

If I programmatically move the servo from 25 degrees to 170 degrees slowly then there is no reset but it is OK if I use webserial to move at speed 60. This ties in with the Asynch_TCP error reported by Arduino serial monitor and must be due to servoEasing code not yielding to a task associated with AsyncTCP which during a servo movement so the watchdog timer event is triggered because it is stating there waiting to respond to the tcp request. The only reason why the yield()is not working is because yield() nor delay() work during a callback or when used in an ISR. I must admit to having difficulty following your code (I am only a hobby programmer) so I so not know if what part of your code is in and ISR and part of a callback.

I suspect that the slow response is again due to the asyncTCP but any delay is not enough to cause the watchdog to trigger and hence a reset.

Note yield() only yields to higher priority tasks and the watchdog timer is a low priority task. delay() yields to all tasks including the watch dog timer that triggers the watchdog event.

[ArminJo]

Note yield() only yields to higher priority tasks and the watchdog timer is a low priority task. delay() yields to all tasks including the watch dog timer that triggers the watchdog event.

Thanks for this information 👍
Reset can eventually avoided by inserting esp_task_wdt_reset(); instead of (or additional to?) yield at line 2413.

[sjgcornish]

That did not work, however I think I have solved the problem, more testing required but the problem was partially mine of the making. My WebSerial.msgCallback(recvMsg) was taking to long and the delay combined with the delay in processing slow long servo movements tripped the watch dog time and it appears for the sometime slow response. I modified the recvMsg callback and put all the code in the main loop.
This whole process has been an education for me.

[sjgcornish]

This makes interesting reading https://www.reddit.com/r/esp32/comments/w2q3r7/esp32_unexplained_slowness_with_simple_iteration/

…

On Mon, 5 Dec 2022 at 08:37, Armin ***@***.***> wrote: Typo! It is 2413. Function updateAllServos() — Reply to this email directly, view it on GitHub <#73 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AH46SWTJKUW42AUY7XEFT2LWLWSTBANCNFSM6AAAAAASS26CEY> . You are receiving this because you authored the thread.Message ID: ***@***.***>