tune.filter

Tests the teleop filter stack offline, with no robot or VR headset. The tool takes a clean joint signal, adds known noise, and replays it through the same smoothing chain and rates teleop uses. You get the same result every time for the same settings. The clean signal is a synthetic sine or a saved reference motion. To run the same test on the real arms, use tune.motion --noise --filter. The diagnostics dashboard runs that version. The teleop pipeline is:

VR stream ──> pose low-pass ──> IK solver ──> EMA ──> trapezoid
   ▲ network noise enters here     ▲ IK noise is created here

Noise sources

--noise picks what to add. Each source enters the pipeline where it would on the robot, so you can test them one at a time.

Mode	Where it enters	What it adds
`network`	Before the pose low-pass, so every filter sees it	Jitter (small noise on every sample), outliers (single samples that jump away) and stalls (the stream freezes, then jumps to catch up)
`ik`	After the low-pass, where the IK solver sits, so only the EMA and trapezoid see it	Churn (3–20 Hz restless motion on each joint) and jumps (one joint switches to another solution for 0.3 to 1 s, then returns)
`combined`	Both places	Both sets of noise (default)

The output is scored per joint against the clean signal. The noise is seeded, and the network and IK noise use separate seeds, so the same seed gives the same network noise whether or not IK noise is added. To compare filter settings, change one (for example --cutoff), rerun, and compare the scores.

Flag	Description
`--noise MODE`	`network`, `ik` or `combined` (default: combined). Network settings are ignored in `ik` mode, and IK settings in `network` mode.
`--motion NAME`	Use a saved reference motion as the clean signal, all 14 joints (default: a synthetic sine)
`--duration S` / `--amp DEG` / `--freq HZ`	Sine length, amplitude and frequency (defaults: 10 s, 15°, 0.5 Hz)
`--jitter DEG`	Network: noise RMS on every sample (default: 0.3°; 0 turns it off)
`--outlier-rate /S` / `--outlier-amp DEG`	Network: how often outliers occur and how far they jump (defaults: 0.5/s, 10°)
`--stall-rate /S` / `--stall-ms MS`	Network: how often stalls occur and how long they last (defaults: 0.5/s, 150 ms)
`--ik-churn DEG`	IK: churn RMS per joint (default: 0.2°; 0 turns it off)
`--ik-jump-rate /S` / `--ik-jump-amp DEG`	IK: how often jumps occur and how large they are (defaults: 0.2/s, 3°)
`--cutoff HZ`	Pose low-pass cutoff (default: teleop’s `pose_cutoff`)
`--seed N`	Noise seed. The same seed gives the same noisy stream (default: 0).
`--label TEXT`	Note stored on the run
`--save-run`	Save the time series and scores under `~/.almond/diagnostics/tuning/` for the diagnostics dashboard

axol tune.filter --noise network --save-run                  # wifi artifacts only
axol tune.filter --noise ik --save-run                       # solver artifacts only
axol tune.filter --save-run                                  # both combined
axol tune.filter --noise network --stall-ms 300 --save-run   # harsher wifi
axol tune.filter --noise ik --ik-churn 0.01 --save-run       # a very restless solver
axol tune.filter --motion reach-and-place --save-run         # a real motion, all joints
axol tune.filter --cutoff 1.5 --label "half cutoff" --save-run

Reading the scorecard

Column	Meaning
`in RMS °`	Error the noise added to the input, against the clean signal
`out RMS °`	Error left after the filters, including their delay on a moving signal
`lagfree °`	`out RMS` with the delay removed. This is the noise the filters left in.
`lag ms`	Delay the filters add
`jitter ×`	Share of the 3–15 Hz error that got through. Well below 1 means the filters removed the jitter you would feel.
`peak °`	Largest error against the clean signal, which shows how much of an outlier got through
`accel`	Peak output acceleration (°/s²). It must stay at or under teleop’s configured limit, so outliers and stall catch-ups can’t jerk the arm.

During a stall the clean signal keeps moving while the stream is frozen, so error during a stall is expected. Judge the filters on the catch-up: the arm should resume without overshoot or a spike in acceleration.

Get Started

Operations

Mantis

Remote Teleop

Web Interfaces

Advanced

tune.filter

Noise sources

Reading the scorecard

​Noise sources

​Reading the scorecard

Noise sources

Reading the scorecard