Commit e8e9629
committed
WIP H1: stiffness 1.0->50.0 / damping 0.1->5.0 diagnostic
Diagnostic probe to discriminate two hypotheses for the Newton ~10x reward
gap on Shadow-Hand-Over MAPPO at the develop pin:
- If reward jumps sharply (e.g. mean from 27 -> 100+), the actuator wiring
works and the baseline gains were just too low; sweep for sweet spot.
- If reward stays ~27, the wiring is broken (e.g. Newton issue #2585
imports MjcActuators as CTRL_DIRECT, fixed in PR #2598 which is 51
commits past our pin a27277ed). Switch to IdealPDActuatorCfg or bump
Newton.
Will revert/replace this commit once the diagnostic is interpreted.1 parent ccac624 commit e8e9629
1 file changed
Lines changed: 12 additions & 8 deletions
Lines changed: 12 additions & 8 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
180 | 180 | | |
181 | 181 | | |
182 | 182 | | |
| 183 | + | |
| 184 | + | |
| 185 | + | |
| 186 | + | |
183 | 187 | | |
184 | | - | |
185 | | - | |
186 | | - | |
187 | | - | |
| 188 | + | |
| 189 | + | |
| 190 | + | |
| 191 | + | |
188 | 192 | | |
189 | 193 | | |
190 | | - | |
191 | | - | |
192 | | - | |
193 | | - | |
| 194 | + | |
| 195 | + | |
| 196 | + | |
| 197 | + | |
194 | 198 | | |
195 | 199 | | |
196 | 200 | | |
| |||
0 commit comments