chenlin/Mujoco_WASM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #612 from qleonardolp:patch-1

Browse Source 
PiperOrigin-RevId: 492705106
Change-Id: I2265892aeca413e221a2cbb2244021c74afefd58
This commit is contained in:
Copybara-Service
2022-12-03 10:49:22 -08:00
parent 0846f38c82 ab56a394bc
commit dc212c276d
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
doc/computation.rst
+1 -1
View File
@@ -71,7 +71,7 @@ and the resulting equations of motion are difficult to characterize. To be fair,
worst-case performance and does not mean that solving the LCP quickly is impossible in practice. Still, convex worst-case performance and does not mean that solving the LCP quickly is impossible in practice. Still, convex
optimization has well-established advantages. In MuJoCo we have observed that for typical robotic models, 10 sweeps of a optimization has well-established advantages. In MuJoCo we have observed that for typical robotic models, 10 sweeps of a
projected Gauss-Seidel method (PGS) yield solutions which for practical purposes are indistinguishable from the global projected Gauss-Seidel method (PGS) yield solutions which for practical purposes are indistinguishable from the global
minimum. Of course there are problems that are much harder to solve numerically, even though the are convex, and for minimum. Of course there are problems that are much harder to solve numerically, even though they are convex, and for
such problems we have conjugate gradient solvers with higher-order convergence. such problems we have conjugate gradient solvers with higher-order convergence.
The requirements for computational efficiency are different depending on the use case. If all we need is real-time The requirements for computational efficiency are different depending on the use case. If all we need is real-time