Principle:Google deepmind Mujoco Quasi Random Sequences
| Knowledge Sources | |
|---|---|
| Domains | Numerical_Methods, Sampling |
| Last Updated | 2026-02-15 06:00 GMT |
Overview
Low-discrepancy sequence generation for quasi-random sampling with better uniformity than pseudo-random numbers.
Description
Quasi-Random Sequences, specifically the Halton sequence, provide a deterministic sequence of numbers in [0, 1) with low discrepancy — meaning the points are more evenly distributed than pseudo-random numbers. In MuJoCo, the Halton sequence is used for generating noise signals in benchmarking (Ornstein-Uhlenbeck control noise) and other sampling tasks where uniform coverage of the input space is important.
Usage
Use when generating reproducible, uniformly-distributed sample sequences for benchmarking, control noise generation, or initialization of simulation states.
Theoretical Basis
The Halton sequence in base b is defined by reversing the base-b representation of the index:
Where are the digits of i in base b. For multi-dimensional sequences, different prime bases are used for each dimension (2, 3, 5, 7, ...).