Little IMU demo #1150
Replies: 5 comments 25 replies
-
|
Alas, the .mp4 link leads me to this: |
Beta Was this translation helpful? Give feedback.
-
|
It's my fat-finger-everything day 😢 |
Beta Was this translation helpful? Give feedback.
-
|
Source code is available at:
All these are work-in-progress! |
Beta Was this translation helpful? Give feedback.
-
|
I'll miss the Worker. The ESP32 has two cores and it is great to use both when it helps. But, it is a bit more complicated so if it isn't necessary, you can hold that knowledge in reserve until it is needed. For your timer you can take advantage of the fact that the repeating timer is drift free. If you schedule a repeating timer with an interval of 1 ms but do more than 1 ms of work, it will give the behavior you want -- a non-blocking event loop that calls you back without delay: Timer.repeat(()=> {
}, 1);I like the option you propose to have the sensor driver return either the latest reading or the oldest in the FIFO. Does the sensor FIFO have a time stamp of some sort so you can return an accurate time stamp for the readings?
Basically... but Passing the device global to the worker succeeds because XS is able to marshall preloaded functions across the worker boundary. That would never work in a browser. But. there's no need for it here because you can just access |
Beta Was this translation helpful? Give feedback.
-
Complicated... The chip can provide timestamps but I'm finding them useless here. The batch() function calculates and provides them in each sample. The time-stamps are based on the fact that the samples in the fifo are collected at a fixed interval and the assumption that if the fifo is not full the first sample was inserted one interval after the last sample previously retrieved. It's a bit inexact, not least because the chip's clock can drift quite a bit (+/-1%). |
Beta Was this translation helpful? Give feedback.
-
|
Time is never easy. And the Bosch sensors often require some black magic to get everything working. I ask because time stamping of sample data is a popular topic in TC53. You can find the results of that in the timing notes in the standard and TR/110. Not critical for your uses, but interesting to try to get right for the more general case. |
Beta Was this translation helpful? Give feedback.
-
|
Yet another thought... One of the pieces in the demo code is the imu.ts module. It combines the raw bmi160 samples and the fusion library to provide processed attitude information. There's no reason this IMU module couldn't use the ECMA-419 Sensor pattern itself! |
Beta Was this translation helpful? Give feedback.
-
|
I like the idea. If the API is a comfortable enough fit, it makes sense to re-use. |
Beta Was this translation helpful? Give feedback.
-
|
We bought a BMI160 sensor and set this up at the Moddable office today. Good news. It works! 🎉It feels just as magical and responsive to use at it looks in your video.
A few notes:
|
Beta Was this translation helpful? Give feedback.
-
|
Yay! Thanks for the notes, I'll fix & good news about the flickering! BTW: the camera angle was chosen to approximate the perspective in the video (which is also the one I had while developing). If you tack the device onto the moddable-two you may want to change the way things are represented... |
Beta Was this translation helpful? Give feedback.
-
Thanks. We did notice. It is awkward to use connected to the device. We're adding a ribbon cable. Forgot to mention, you seem to be using a version of xs.d.ts that defines I have an idea for how to fix the GC problem that should be general and efficient. Just need to find some time to implement it. |
Beta Was this translation helpful? Give feedback.
-
|
Someone didn't do any release engineering... :-) |
Beta Was this translation helpful? Give feedback.
-
|
We do accept PRs. 😉 |
Beta Was this translation helpful? Give feedback.
-
|
I updated the BMI160 driver, broke out an AHRS library, tweaked the Fusion library, and updated the main app:
The worker performs the AHRS stuff, the main does the 3D projection and the drawing. One CPU is 98% busy the other 22%. It would probably be yet faster if the 3D calcs were moved to the worker. But I fear that's beyond my patience now 😆. I patterned the AHRS library after the Sensor class pattern but I don't like it. As it stands the information to construct the IMU device is passed in ECMA-419 style and the AHRS class constructs the device. But all it really does with it is to call (I didn't do a fresh install based on what's in the repo, please let me know if I forgot something again...) |
Beta Was this translation helpful? Give feedback.
-
|
Haha, you should have bought the 3-pack! 😂 BTW: what do the two numbers in the profile mean? self and total-with-children? Is that the same in top-down vs. bottom up mode? Might be nice to write a section in the xsbug.md? |
Beta Was this translation helpful? Give feedback.
-
|
@tve, I don't know if you saw these blog posts, but they have some detail on the profiler. |
Beta Was this translation helpful? Give feedback.
-
|
I did see them but I didn't see any description of what the numbers mean... |
Beta Was this translation helpful? Give feedback.
-
|
I've never met a profiler that didn't take me some time to learn to interpret the results. The XS profiler follows the approach of other JavaScript profilers to try to minimize the learning curve (we tested by exporting XS profile data to view other tools). A simple test program often helps me understand the profiler output. import Timer from "timer"
function leaf() {
for (let i = 0; i < 100; i++)
i * i * i;
}
function runner() {
for (let j = 0; j < 100; j++)
leaf();
}
Timer.repeat(runner, 1);Running that in the simulator gives me this: 
That makes it clear that the left column is time in the function itself and the right column is elapsed time in the function (function + children). Note that XS implements a sampling profiler, so the results are likely not identical across runs and can miss functions that don't execute for long and are called infrequently. As a JavaScript profiler, it doesn't have visibility into the native host code. All that said, it is incredibly useful for quickly identifying hotspots and evaluating the impact of optimizations. |
Beta Was this translation helpful? Give feedback.
-
|
Detailed documentation on the XS Profiler implementation is in this repository in XS Profiler.md. The Viewers section explains how the numbers displayed are calculated. |
Beta Was this translation helpful? Give feedback.
-
|
I had some time to rework the Here are the sources: hack3d.d.ts// Based on https://github.com/jordoreed/home-grown-graphics/blob/main/index.html
// By Jordan Reed
declare module "hack3d" {
function randomFloat(min: number, max: number): number
function toRadians(degrees: number): number
interface Vec4 {
x: number
y: number
z: number
w: number
}
class Matrix4 {
m: Float64Array
constructor(...args: number[])
constructor(arr: Float64Array)
constructor(...args: number[] | Float64Array[])
multiply(r: Matrix4): Matrix4
multiplyVec4(v: Vec4): Vec4
static identity(): Matrix4
static perspective(l: number, r: number, b: number, t: number, n: number, f: number): Matrix4
static perspectiveAspectRatio(aspectRatio: number, fov: number, n: number, f: number): Matrix4
static translate(x: number, y: number, z: number): Matrix4
static scale(x: number, y: number, z: number): Matrix4
// pitch
static rotateX(degrees: number): Matrix4
// yaw
static rotateY(degrees: number): Matrix4
// roll
static rotateZ(degrees: number): Matrix4
}
}hack3d.js// Based on https://github.com/jordoreed/home-grown-graphics/blob/main/index.html
// By Jordan Reed
export function randomFloat(min, max) {
return Math.random() * (max - min) + min
}
export function toRadians(degrees) {
return (degrees * Math.PI) / 180
}
export class Matrix4 {
constructor() @ "xs_Matrix"
multiply(r) @ "xs_matrix_multiply"
multiplyVec4(v) @ "xs_matrix_multiplyVec4"
static identity() {
return new Matrix4(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1)
}
static perspective(l, r, b, t, n, f) {
return new Matrix4(
(2 * n) / (r - l),
0,
(r + l) / (r - l),
0,
0,
(2 * n) / (t - b),
(t + b) / (t - b),
0,
0,
0,
-(f + n) / (f - n),
(-2 * f * n) / (f - n),
0,
0,
-1,
0
)
}
static perspectiveAspectRatio(aspectRatio, fov, n, f) {
const halfH = Math.tan(toRadians(fov * 0.5)) * n
const halfW = aspectRatio * halfH
return Matrix4.perspective(-halfW, halfW, -halfH, halfH, n, f)
}
static translate(x, y, z) {
return new Matrix4(1, 0, 0, x, 0, 1, 0, y, 0, 0, 1, z, 0, 0, 0, 1)
}
static scale(x, y, z) {
return new Matrix4(x, 0, 0, 0, 0, y, 0, 0, 0, 0, z, 0, 0, 0, 0, 1)
}
// pitch
static rotateX(degrees) {
const radians = toRadians(degrees)
const c = Math.cos(radians)
const s = Math.sin(radians)
return new Matrix4(1, 0, 0, 0, 0, c, -s, 0, 0, s, c, 0, 0, 0, 0, 1)
}
// yaw
static rotateY(degrees) {
const radians = toRadians(degrees)
const c = Math.cos(radians)
const s = Math.sin(radians)
return new Matrix4(c, 0, s, 0, 0, 1, 0, 0, -s, 0, c, 0, 0, 0, 0, 1)
}
// roll
static rotateZ(degrees) {
const radians = toRadians(degrees)
const c = Math.cos(radians)
const s = Math.sin(radians)
return new Matrix4(c, -s, 0, 0, s, c, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1)
}
}hack3d.c#include "xsmc.h"
#include "mc.xs.h" // for xsID_ values
#include "xsHost.h"
void xs_Matrix(xsMachine *the)
{
if ((1 == xsmcArgc) && xsmcTest(xsArg(0))) // instanceof Float64Array and length === 16
xsmcSet(xsThis, xsID_m, xsArg(0));
else if (16 == xsmcArgc) { // typeof first argument is number
xsNumberValue d[16], *r;
xsUnsignedValue rSize;
xsmcVars(1);
xsmcSetInteger(xsVar(0), 16);
xsmcNew(xsVar(0), xsGlobal, xsID_Float64Array, &xsVar(0), NULL);
xsmcSet(xsThis, xsID_m, xsVar(0));
for (int i = 0; i < 16; i++)
d[i] = xsmcToNumber(xsArg(i));
xsmcGet(xsVar(0), xsVar(0), xsID_buffer);
xsmcGetBufferWritable(xsVar(0), (void **)&r, &rSize);
if (128 != rSize)
xsUnknownError("unexpected");
c_memmove(r, d, rSize);
}
else
xsUnknownError("Invalid arguments");
}
void xs_matrix_multiply(xsMachine *the)
{
xsNumberValue *m, *r, *result;
xsUnsignedValue mSize, rSize, resultSize;
xsSlot n;
xsmcVars(3);
xsmcSetInteger(n, 16);
xsmcNew(xsVar(0), xsGlobal, xsID_Float64Array, &n, NULL);
xsmcNew(xsResult, xsThis, xsID_constructor, &xsVar(0), NULL);
xsmcGet(xsVar(0), xsThis, xsID_m);
xsmcGet(xsVar(0), xsVar(0), xsID_buffer);
xsmcGet(xsVar(1), xsArg(0), xsID_m);
xsmcGet(xsVar(1), xsVar(1), xsID_buffer);
xsmcGet(xsVar(2), xsResult, xsID_m);
xsmcGet(xsVar(2), xsVar(2), xsID_buffer);
xsmcGetBufferReadable(xsVar(0), (void **)&m, &mSize);
xsmcGetBufferReadable(xsVar(1), (void **)&r, &rSize);
xsmcGetBufferWritable(xsVar(2), (void **)&result, &resultSize);
if ((mSize != 128) || (rSize != 128) || (resultSize != 128))
xsUnknownError("invalid input");
result[0] = m[0] * r[0] + m[1] * r[4] + m[2] * r[8] + m[3] * r[12];
result[1] = m[0] * r[1] + m[1] * r[5] + m[2] * r[9] + m[3] * r[13];
result[2] = m[0] * r[2] + m[1] * r[6] + m[2] * r[10] + m[3] * r[14];
result[3] = m[0] * r[3] + m[1] * r[7] + m[2] * r[11] + m[3] * r[15];
result[4] = m[4] * r[0] + m[5] * r[4] + m[6] * r[8] + m[7] * r[12];
result[5] = m[4] * r[1] + m[5] * r[5] + m[6] * r[9] + m[7] * r[13];
result[6] = m[4] * r[2] + m[5] * r[6] + m[6] * r[10] + m[7] * r[14];
result[7] = m[4] * r[3] + m[5] * r[7] + m[6] * r[11] + m[7] * r[15];
result[8] = m[8] * r[0] + m[9] * r[4] + m[10] * r[8] + m[11] * r[12];
result[9] = m[8] * r[1] + m[9] * r[5] + m[10] * r[9] + m[11] * r[13];
result[10] = m[8] * r[2] + m[9] * r[6] + m[10] * r[10] + m[11] * r[14];
result[11] = m[8] * r[3] + m[9] * r[7] + m[10] * r[11] + m[11] * r[15];
result[12] = m[12] * r[0] + m[13] * r[4] + m[14] * r[8] + m[15] * r[12];
result[13] = m[12] * r[1] + m[13] * r[5] + m[14] * r[9] + m[15] * r[13];
result[14] = m[12] * r[2] + m[13] * r[6] + m[14] * r[10] + m[15] * r[14];
result[15] = m[12] * r[3] + m[13] * r[7] + m[14] * r[11] + m[15] * r[15];
}
void xs_matrix_multiplyVec4(xsMachine *the)
{
xsNumberValue wIn, xIn, yIn, zIn;
xsNumberValue wOut, xOut, yOut, zOut;
xsNumberValue *m;
xsUnsignedValue mSize;
xsSlot n;
xsmcVars(1);
xsmcGet(n, xsArg(0), xsID_w); wIn = xsmcToNumber(n);
xsmcGet(n, xsArg(0), xsID_x); xIn = xsmcToNumber(n);
xsmcGet(n, xsArg(0), xsID_y); yIn = xsmcToNumber(n);
xsmcGet(n, xsArg(0), xsID_z); zIn = xsmcToNumber(n);
xsmcGet(xsVar(0), xsThis, xsID_m);
xsmcGet(xsVar(0), xsVar(0), xsID_buffer);
xsmcGetBufferReadable(xsVar(0), (void **)&m, &mSize);
if (128 != mSize)
xsUnknownError("invalid");
xOut = m[0] * xIn + m[1] * yIn + m[2] * zIn + m[3] * wIn;
yOut = m[4] * xIn + m[5] * yIn + m[6] * zIn + m[7] * wIn;
zOut = m[8] * xIn + m[9] * yIn + m[10] * zIn + m[11] * wIn;
wOut = m[12] * xIn + m[13] * yIn + m[14] * zIn + m[15] * wIn;
xsmcSetNewObject(xsResult);
xsmcSetNumber(n, xOut); xsmcSet(xsResult, xsID_x, n);
xsmcSetNumber(n, yOut); xsmcSet(xsResult, xsID_y, n);
xsmcSetNumber(n, zOut); xsmcSet(xsResult, xsID_z, n);
xsmcSetNumber(n, wOut); xsmcSet(xsResult, xsID_w, n);
}I also tweaked function calcBounds(vertices: Vec4[]): Bounds {
let x1 = vertices[0].x, y1 = vertices[0].y, x2 = vertices[0].x, y2 = vertices[0].y;
for (let i = 0, length = vertices.length; i < length; i++) {
const v = vertices[i];
// +/-1 is due to width of stroke in Outline.stroke()
if (v.x <= x1) x1 = v.x - 1
if (v.x >= x2) x2 = v.x + 1
if (v.y <= y1) y1 = v.y - 1
if (v.y >= y2) y2 = v.y + 1
}
return {x1, x2, y1, y2}
} |
Beta Was this translation helpful? Give feedback.
-
|
@tve – curious if you had a chance to give this a try? If you prefer, I could make a PR to your repo. |
Beta Was this translation helpful? Give feedback.
-
|
A PR is definitely appreciated, but I'm scrambling for a project (that uses the IMU but not the 3D viz) so this is going to be stuck in the queue for a few more days... |
Beta Was this translation helpful? Give feedback.
-
|
PR is now there. tve/modtve#1 FWIW – one more change is needed to make this build without changes: the fusion library would either need to be included in modtve/lib/fusion or added to the manifest as a Git import. I wasn't sure about your preference so I just left it as-is. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
I'm working on an attitude sensor that tracks the roll & pitch of an ocean racing kayak (surfski) and in order to get a sense of the data quality I put together a little graphics demo using a moddable two. It's now showing signs of life :-).
Stuff used:
Code architecture:
Short demo video: https://tve.s3.amazonaws.com/public/imu-demo-20230610-1300.mp4
Getting 16fps rendering is pretty darn amazing, Poco rocks!
Beta Was this translation helpful? Give feedback.
All reactions