MPPI documentation inconsistency about docs.nav2.org HOT 1 OPEN

That is general advice which is true. The discrepancy larger will impact performance the larger these values diverge. This advice is especially pointed to people trying to make it work at 10-15-20hz where the model_dts there are so huge the controller isn't super stable.

It still applies to the situation you rightly point out. But that was a tactical trade off between making the discrepancy low, yet having a high enough prediction time (model_dt * number of points), and a small enough number of points to work on a broad range of compute platforms out of the box.

You'd be right to say that if you're tuning and you have CPU you can spend, reducing that to 1/freq is a good potential move. The value add of much smaller after you've hit ~0.04 & at or lower than 1/freq is marginal (if any) and its better to optimize for other things like longer time horizons or more samples in your batches -- unless in exceptional situations

from docs.nav2.org.