T35 issue with floats/int32s conversion - maybe?

[mjs513]

Hi all,

I have been trying to implement a waypoint navigation method that will generate intermediate flyby waypoints that is based on what was implemented in Ardupilot. What I did was to take the sections needed to generate the points and put them in a library. That is seems to work ok at least everything compiles and runs. But in my debugging process to test the code I am running into a problem that I think is a conversion issue (don't know it that's it or something else in the code). Anyway I will try and walk through the code pieces to demonstrate what I am talking about. Again I am using a T3.5 (also tried T3.2 with same results), TD1.40 and IDE1.8.5 on a Windows 10 machine.

first I setting up a series of test points like so:

Vector2f test_points1[] = {
    Vector2f(-117.041076, 32.844287),
    Vector2f(-117.037579, 32.844269),
    Vector2f(-117.038373, 32.841664),
    Vector2f(-117.035937, 32.840222),
    Vector2f(-117.033448, 32.842854),
    Vector2f(-117.032762, 32.845774),
    Vector2f(-117.039585, 32.846820)
};

vector2f comes from ardupilot vector2 library and is based on templates. This is a portion of the lib:
vector2.h:

#ifndef VECTOR2_H
#define VECTOR2_H

#include <math.h>

template <typename T>
struct Vector2
{
    T x, y;

    // trivial ctor
    Vector2<T>() {
        x = y = 0;
    }

    // setting ctor
    Vector2<T>(const T x0, const T y0) : x(x0), y(y0) {
    }

    // function call operator
    void operator ()(const T x0, const T y0)
    {
        x= x0; y= y0;
    }

....
typedef Vector2<int16_t>        Vector2i;
typedef Vector2<uint16_t>       Vector2ui;
typedef Vector2<int32_t>        Vector2l;
typedef Vector2<uint32_t>       Vector2ul;
typedef Vector2<float>          Vector2f;

In the ino file I have:

struct Location loc;

static struct Location location_from_point(Vector2f pt)
{
    loc.lat = pt.x * 1.0e7f;
    loc.lng = pt.y * 1.0e7f;
    loc.alt = 0;
    return loc;
}

stuct Location is defined in location.h as:

struct Location
{
    long lat = 0;
    long lng = 0;
    long alt = 0;
};

When I do some debug prints:

test_points1[0].x  = -117.04107666  when it should read -117.041076, very small delta - probably ok

setting

struct Location Aloc = location_from_point(test_points1[i]);

results in:

A(1)=-1170410752; A(2)=328442880;

which should be

-1170410760, 328442870

I am attaching everything in a couple of zip files - just in case.
View attachment Flyby_Waypoint.zip View attachment AP_Math_freeimu.zip

Any help would be appreciated

thanks
Mike

 

[manitou]

float only has precision of 6 decimal digits, so for higher precision you may need to switch to double. double will make computations slower, since T3.5 can do float's in hardware.

https://en.wikipedia.org/wiki/Single-precision_floating-point_format

 

[mjs513]

thanks manitou. That seemed to be the problem. As a quick test I changed the 1e7f to 1e6f as a quick test and things got much better - at least now I can continue debugging. At some point I will have to add double to the vectors and change the rest of the libs as well but at least for now is good enough.

UPDATE 11/26/17
I changed to doubles and still running into something that is confusing me between doubles and floats so I ran a little test sketch:

void setup() {
  Serial.begin(115200);
  delay(5000);
  // put your setup code here, to run once:
  float x = -73.814664;
  double y = -73.814664;
  int32_t z = -73814664;

  Serial.printf("x=%f;\n",x);
  Serial.printf("y=%lf;\n",y);
  Serial.printf("z=%li; ",z);
  Serial.println();
}

void loop() {}

The results showed no difference in precision between double and float when I printed it out? I am confused now:
x=-73.814667;
y=-73.814667;
z=-73814664;

Thanks
Mike

 

[manitou]

ultimately you are at the mercy of how Serial.printf handles floats/doubles. For higher precision output, look at sketch at end of the following thread
https://forum.pjrc.com/threads/21479-Teensy-3-0-and-Arduino-precision-of-double

also teensy compiler for T3.5 coerces floating point constants to float, for double constant you need an L, e.g.
double y = -73.814664L;

 

[mjs513]

the issue seemed to boil down to your second point. I added "L" to all the double constants and that seemed to do the trick. I never realized that was the case.

As for your first point I used "%lf" for double precision values in printf and it printed the correct values - I did compare it to the double2s function in the referenced thread and they were exactly the same.

Thanks again. Now for some more debugging.
Mike