Card: results hold under benchmark 0.2.0's joint speed limit
Browse files
README.md
CHANGED
|
@@ -71,6 +71,12 @@ in the scene (0.31). When two objects swap places, or the instruction
|
|
| 71 |
names another object, its motion changes as much as the teacher's does
|
| 72 |
(1.00). So it uses both the cameras and the words.
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
**How the checkpoint was chosen.** The training run checked itself every
|
| 75 |
2,500 steps on 100 held-out scenes, and those turned out to be the first
|
| 76 |
100 of the 300 above (seeds 9,010,000 to 9,010,099); v0.1.0 was picked
|
|
|
|
| 71 |
names another object, its motion changes as much as the teacher's does
|
| 72 |
(1.00). So it uses both the cameras and the words.
|
| 73 |
|
| 74 |
+
These results are the same under the benchmark's version 0.2.0, which adds a
|
| 75 |
+
joint speed limit (no joint faster than 0.12 rad per step): it slowed 0.13% of
|
| 76 |
+
AMP Tadpole's steps, and no clause came out differently. AMP Tadpole takes 1.13
|
| 77 |
+
times the teacher's steps on the clauses both complete, and its path is 1.96
|
| 78 |
+
times a straight line (the teacher's: 1.90).
|
| 79 |
+
|
| 80 |
**How the checkpoint was chosen.** The training run checked itself every
|
| 81 |
2,500 steps on 100 held-out scenes, and those turned out to be the first
|
| 82 |
100 of the 300 above (seeds 9,010,000 to 9,010,099); v0.1.0 was picked
|