Multiple inertial support #6
Conversation
|
going to implement outlier detection |
|
Could this PR be split into two parts? I'd like to go ahead with the |
| let mut avg: Angle = Angle::from_radians(0.0); | ||
| // calculate initial average | ||
| drop(success_value.iter().map(|angle| avg += angle.to_owned())); | ||
| avg /= N as f64; |
There was a problem hiding this comment.
Why isn't this just let avg = success_value.iter().reduce(|acc, angle| acc + angle);
There was a problem hiding this comment.
Wait no, needs slightly more logic, but should be fine with a fold
| /// 7.5 radian degree tolerance | ||
| const MAX_TOLERANCE_RADIANS: f64 = 7.5; | ||
| const MAX_TOLERANCE_G: f64 = 1.5; | ||
| impl<const N: usize> Gyro for [InertialSensor; N] { |
There was a problem hiding this comment.
Could this potentially be extra generalized to all G: Gyro (i.e impl<G: Gyro, const N: usize> Gyro for [G; N] {...})
There was a problem hiding this comment.
Don't think this should be any more general due to there being hardcoded tolerance values here.
| let mut avg: f64 = 0.0; | ||
| // calculate initial average | ||
| drop(success_value.iter().map(|vel| avg += vel.to_owned())); | ||
| avg /= N as f64; |
There was a problem hiding this comment.
Can be converted to a fold/reduce
| let mut variance: f64 = 0.0; | ||
| drop(success_value.iter().map(|vel| variance += vel.to_owned())); | ||
| variance /= N as f64 - 1.0; |
There was a problem hiding this comment.
Can be converted to a fold/reduce
| let mut ignored: u8 = 0; | ||
| for (index, heading) in success_value.into_iter().enumerate() { | ||
| let z_score = (heading - avg) / standard_deviation; | ||
| if f64::abs(z_score) >= MAX_TOLERANCE_G { | ||
| success_value[index] = 0.0; | ||
| ignored += 1; | ||
| } | ||
| } | ||
| // calculate new average | ||
| drop(success_value.iter().map(|vel| avg += vel.to_owned())); | ||
| avg /= (N - ignored as usize) as f64; |
There was a problem hiding this comment.
This is also broken; avg is not set to zero, and will add to the existing avg value. Can probably benefit from a filter_map().fold()
| // calculate new average | ||
| drop(success_value.iter().map(|angle| avg += angle.to_owned())); | ||
| avg /= (N - ignored as usize) as f64; |
There was a problem hiding this comment.
suffers from the same bug as [InertialSensor; N]::angular_velocity
| let mut variance: f64 = 0.0; | ||
| drop( | ||
| success_value | ||
| .iter() | ||
| .map(|angle| variance += angle.to_owned().as_radians()), | ||
| ); | ||
| variance /= N as f64 - 1.0; |
| /// Method of retreiving heading | ||
| fn heading(&mut self) -> Result<Angle, Self::Error>; | ||
| /// get raw angular velocity readings | ||
| fn angular_velocity(&mut self) -> Result<f64, Self::Error>; |
There was a problem hiding this comment.
Any reason these are required to be mutable?
There was a problem hiding this comment.
like i could be just dumb and not realize that IMU functions are mutable, but it doesn't make sense imho, as state reading shouldn't be mutable.
There was a problem hiding this comment.
The trait takes a mutable reference to self because you could want to modify some internal struct values, kinda just took the suggestion of tropical. Makes sense to me as a feature of flexibility
wijiler
force-pushed
the
multiple-inertial
branch
from
0d2cd78 to
fde3afe
Compare
Adds support for multiple inertials, allowing for reduced noise, and more accurate heading readings. Overall a very useful feature missing from most motion libraries
Example, module and library both build
Negatives: