Problem solved! After triplechecking the batteries, the servo wiring, SW3, running ' moduleTests.ino', and more, I started suspecting the instinct data in the EEPROM as being corrupted. However, updating the instincts data in 'WriteInstinct.ino' was no longer possible. The option on the screen for writing instincts had appeared only the first time that I ran that program and then no more. But with help from my friends the program code in 'WriteInstinct.ino' was amended (after line 219 new code was added to empty the read buffer), to make the option 'Update Instincts? (Y/n)' show again. So then the instincts were freshly written to the external I2C EEPROM. And with that the cat started walking properly, no more 'chaos'.

@Rongzhong Li The leg movements in the 'moduleTests.inio' show walking patterns, but the front legs seem a little bit high for touching ground, that's what worried me. You raise an interesting point about the motion sensor. In Video_2 the cat is indeed running 'Nybble.ino', but the tethering doesn't make a difference. Holding it steadily up by hand gives the same result. If you look at the second video from the top of this thread (confusingly called 'video 2') you'll see the very same moves with the cat standing on a table, also running 'Nybble.ino', also after being properly initialized and calibrated. If the MPU plays a role it goes for both cases, whether hanging loose or fixed.

No, the yellow light doesn't blink, it's steady. I even put two sets of identical batteries (https://eu.nkon.nl/vapcell-icr14500-1000mah-3a.html) in parallel and that didn't make a difference. The voltage is approx. 8.2 in rest and 7.9 when legs are moving. That should settle doubts about the power, doesn't it? In 'moduleTests.ino', in the test for the servos, it shows repetitive leg motions as in my Video_1. Are these motions as they _should_ be? The leg motions in my Video_2, when running 'Nybble.ino' are also repetitive (and not chaotic as I first thought), but they are identical for a number of different commands, like sit, walk, stretch and even pee. If the hardware and circuit and also the power are ok, then where to look next?

No, stronger batteries didn't solve the issue. Further investigation led me to the sketch 'moduletests.ino'. The first test 'check servo driver' proved interesting: it showed coordinated leg movements, though at strange angles for the front legs. See video 1 (Sorry for the decapitation of our little friend, but I had to save the head from being hit violently by the front legs...). However I cannot draw any conclusions from this test. 'Nybble.ino' still shows these wild repetitive movements for any commands. Only 'c', 'd' or 'rest', can stop it, and then allways in the same calibration posture. See video 2, There seem to be only two options: 'chaos' or the calibrate posture. Hmm. Forum member leukipp reported similar behavior, (https://www.petoi.com/forum/basic-assembly-and-setup/the-logic-behind-calibration-process), but his/her solution didn't help me, as the servos are already attached according to the Joint Map (doublechecked)

Hi, thanks for your quick reply! I have checked the points you mentioned. SW3 is correct: between BATT and V_S. Nybble is also under power without USB. The blue and yellow leds are on and there is no continous beeping. The flat head of the batteries is no problem (don't ALL unprotected batteries have flat heads?). I put blobs of solder on the contacts of the battery holder. Still, after all, I now have some doubts about the batteries, despite their specs. I have ordered some more powerfull ones: https://eu.nkon.nl/rechargeable/li-ion/14500-16340/vapcell-inr14500-1000mah-10a.html. Let's see...