F/optimize floats and tensors

[artofnothingness]

No description provided.

[SteveMacenski]

I'll do a deep dive tomorrow on this - do you notice this helps with the jitter and/or performance?

[artofnothingness]

I'll do a deep dive tomorrow on this - do you notice this helps with the jitter and/or performance?

looks like it. old benchmarks became 3 times faster 2.2 ms vs 0.7 ms

[SteveMacenski]

Heck yeah! 😄

[SteveMacenski]

The code generally looks good, though I'm seeing a degradation in performance (e.g. robot getting stuck more, its going oddly fast).

Ex.

linear:
  x: 1.1755331754684448
  y: 0.0
  z: 0.0
angular:
  x: 0.0
  y: 0.0
  z: 0.3975425064563751

When the max is supposed to be 0.5. So I think something's not quite right. I just can't seem to find it from a look here.

I see the performance to about 13ms per iteration, which is pretty remarkable (~80hz) however I still see the spikes unfortunately. I don't think that's a reason not to still work on and merge this for the outright performance improvements though. This is still a really significant step forward from that perspective. About a 35% improvement.

I also just tested, and with these changes and after the jitter issue is resolved, we could actually back out the noise generator thread. Now the strides aren't so complex, the noises being generated inline in the main thread only adds ~2ms to the average runtime, which is still well within our desired performance

[SteveMacenski]

LGTM and fixed the prior issue -- curious what that was? I went through your changes and I actually don't see even in the solution. What was it, just for my curiosity (and help tune my future reviewing skills)?

[artofnothingness]

@SteveMacenski i clipped velocities and not controls )