Well, let's try something to see if we can work it out.
1.  Ensure 4 axis are all at the same level.
2.  Check the value in X,Y,Z of GPS location.
3.  Confirm the centre of gravity.

Make sure the motors are level. Even a single degree out of level and it will turn. It's a real down side to the Tarrot frame. On the DJI frames there is no way to turn the motors accidentally. They are all properly aligned.  

Is the gain value too  high?

I have this problem too. So do most people... Sounds like a firmware bug of some sort

freddiebd.iclou Posted at 2015-2-12 02:58
I have this problem too. So do most people... Sounds like a firmware bug of some sort

Have you solved this problem? I think your IMU may need recalibrating?

Nope still not solved. IMU calibrated many times. Ill just have to wait for the new Naza v2 firmware coming out soon

I'm having the same issues. I'm very new to the Hobby, purchased an off-brand Tarot like frame and there is obviously room for error when mounting the motors as they are (Tarot like system). So Yaw Gain needs to be lowered rather than increased (100% down to 90%) and make sure the motor mounts are level. It is kind of a bummer when you spend $1300 on a Hexcopter and you start having issues. It has become pretty awkward to fly at times due to this problem. What it sounds like is it's the frame.

It looks like DJI has no business to think about this problem. Got it on my quad also after updating from 3.18. Before the update everything was ok. I have no response from DJI for quite a while on this issue.
I've tried everything, including downgrading the soft and remounting everything on my frame to eliminate any possible magnetic interference - nothing helped.
Its obviously a FC issue, the drone after yawning flies away to the right or to the left (depends on the direction of the spin), than stops and goes back to the point of the rotation (place where it should be all the time).
Im starting to think that they just let go of naza assistance and got focused on the new P3 and Inspire... even A2 problems are left without an answer...

I had yaw issues when my compass was off a tad.  Naza uses the compass to know its heading but once the craft is moving it is able to get heading info from the gps too which is more reliable than the compass so it's possible Naza is trying to adjust for both so the heading matches gps and compass.  My pixhawk flight controller would yaw on me once moving 6 m/s cuz it would start listening to the heading info it got from the gps and when I slowed down and gps heading wasn't reliable it would use heading info from the compass and yaw again. The issue was the cheap Chinese gps and compass assembly I was using had the arrow on the gps puck in the wrong place. Easy to figure out with pixhawk be because when connected to mission planner the copter was facing north in  my bedroom and mission planner was showing north east so I just rotated the gps on the stand until mission planner showed north like the copter was physically facing and that solved my issue..  With Naza there is no mission planner so you have to guess or use the azimuth info off of iosd to get the degrees Naza thinks it's facing and compare it to how the copter is physically facing and turn Naza gps puck until they line up. No iosd then your guessing..  If it's more than 25 degrees off I've noticed is when the toilet bowling will start...

Hope I helped you guys

Mark its not a toilet bowl, it simply runs away to one side after the yaw, than stops and goes back to the point of the yaw. Looks like the gps is getting set to the point cuple of meters away, thac gets a new (old and proper) position fixed again and flyes back in the right place.

is there any update on this issue it is very frustrating.

I have the same issue here to.

All was working fine, added the iOsd and reflashed everything to make sure all features were picked up and now the qwad yaws to the right and is very erratic.

Spent around £400 of all of the DJI stuff for my custom built qwad and now it looks like it was a waste of money!

Should have gone the tarot route!

DJI, PLEASE SOLVE THIS MASSIVE ISSUE!

I am the author of AeroSIM RC, and I have experienced the behaviour described in this topic with a Naza V1.

I have an explanation, which I discovered when implementing the GPS flight mode in the DJI-F450 drone in the simulator.

It is related to the Integral term of the PIDs that control the position.

The GPS flight mode in the simulator uses two PID controllers: one for X and other for Y axis.
The Integral term of the PID controller is the one than cancels the drift with the wind.

When the user yaws the drone, the Integral terms of the X and Y PID controllers require some time to reach the new value, and in the meantime, they act in the wrong direction.

To solve the drift not found in modern flight controllers, in the latest version v4.61 of AeroSIM RC, I have added an option to the simulator which rotates the X and Y integral terms according to the drone yaw. This completelly solves the issue.