GPS reference frame and position_init_yaw about ethzasl_msf HOT 10 CLOSED

@blackistheanswer we usually set a local level frame of reference (init GPS position and init yaw position). From then on the incoming GPS measurements are converted to metric ECEF and the offset from ECEF to local level is subtracted before the measurement gets applied to the filter. The yaw direction has to be set as the offset w.r.t. north.

from ethzasl_msf.

@simonlynen So basically, would it be prudent to set yaw direction automatically at start from compass measurements?

from ethzasl_msf.

@mhkabir this would be fine if your compass is not influenced by nearby devices or the robot itself. In fact we usually used a very rough initial guess. The filter has a separate transformation which is estimated to adjust any wrong initial guess of this estimate. This way it is fine if you are approx. right.

from ethzasl_msf.

@simonlynen I'm adding GPS measurements to MSF now that I've been successful with IMU+visual SLAM. I am a bit confused about the function of msf_distort. I see in the launch files, sometimes the GPS data is plugged into distort and the distort output is mapped to the core's GPS input, and sometimes just directly.

Can you please briefly tell us.

from ethzasl_msf.

@mhkabir The msf-distort is just for unit-testing and experimentation. It can simulate added delays and sensor-dropouts to the signal. You can either remove the piping through distort entirely or just ignore it.

from ethzasl_msf.

Oh cool :) Thank you.

On Fri, Sep 12, 2014 at 8:41 PM, Simon Lynen [email protected]
wrote:

@mhkabir https://github.com/mhkabir The msf-distort is just for
unit-testing and experimentation. It can simulate added delays and
sensor-dropouts to the signal. You can either remove the piping through
distort entirely or just ignore it.

—
Reply to this email directly or view it on GitHub
#78 (comment).

from ethzasl_msf.

@simonlynen Is core_height the absolute height from ground? If we were to again dynamically initialize this with data from a laser rangefinder and call core_init_height, would it work out? i.e Do we get correct scaling from start in that case? i.e without needing the the filter states to converge by common movement?

Also, if core_init_height is not called, and we call core_init_filter(as I've been doing) without any height data, what sort of movements and time would be needed for the states to converge and get proper scaled output of the pose source?

from ethzasl_msf.

• core_height == absolute height above ground.
• you can call core_init_height at any time as long as you are not using the state estimate at this point in a controller. (The init will reset and initialize the filter).
• If you have an absolute metric measurement of height which is measuring the altitude in the same frame of reference as your SLAM (SVO) then you will have a correct scale estimate from the beginning on.
• The question about the needed movement is a bit difficult to answer. We usually started the SLAM in flight and then set the height to a rough value (+= 0.2m) which was sufficient to start the flight without additional movements. Of course the scale convergence will be slower if you don't have acceleration measurements apart from gravity.
• Absolute-pose means that the pose is used as an absolute measurement. This contrasts with the "relative measurements" that the filter can process as measurements between states. (Which uses the stochastic cloning technique).

from ethzasl_msf.

Thank you so much.

On Fri, Sep 12, 2014 at 9:33 PM, Simon Lynen [email protected]
wrote:

• core_height == absolute height above ground.
• you can call core_init_height at any time as long as you are not
  using the state estimate at this point in a controller. (The init will
  reset and initialize the filter).
• If you have an absolute metric measurement of height which is
  measuring the altitude in the same frame of reference as your SLAM (SVO)
  then you will have a correct scale estimate from the beginning on.
• The question about the needed movement is a bit difficult to answer.
  We usually started the SLAM in flight and then set the height to a rough
  value (+= 0.2m) which was sufficient to start the flight without additional
  movements. Of course the scale convergence will be slower if you don't have
  acceleration measurements apart from gravity.
• Absolute-pose means that the pose is used as an absolute
  measurement. This contrasts with the "relative measurements" that the
  filter can process as measurements between states. (Which uses the
  stochastic cloning technique).

—
Reply to this email directly or view it on GitHub
#78 (comment).

from ethzasl_msf.

Assuming this is resolved.

from ethzasl_msf.