Update results: PushT 98% rh1 / 92% rh5, Reacher 100%, TwoRooms 100%
#2
by nottygian - opened
README.md
CHANGED
|
@@ -25,8 +25,8 @@ LePlanner is an amortized iterative controller that plans through a **frozen JEP
|
|
| 25 |
|
| 26 |
| Environment | Method | Success | Notes |
|
| 27 |
|---|---|---|---|
|
| 28 |
-
| PushT | LePlanner (rh=1) | **
|
| 29 |
-
| PushT | LePlanner (rh=5) | **
|
| 30 |
| PushT | CEM (rh=5) | 90% | baseline search planner |
|
| 31 |
| Reacher | LePlanner (rh=1) | **100%** | 50 episodes, seed 42 |
|
| 32 |
| TwoRooms | LePlanner (rh=1) | **100%** | 50 episodes, seed 42 |
|
|
@@ -97,7 +97,7 @@ results/ # Raw evaluation logs
|
|
| 97 |
|
| 98 |
| File | Description |
|
| 99 |
|---|---|
|
| 100 |
-
| `checkpoints/ablations/ah_hold0.5/controller.pt` | **Headline PushT controller** β
|
| 101 |
| `checkpoints/environment_controllers/reacher/controller.pt` | Reacher controller β 100% |
|
| 102 |
| `checkpoints/environment_controllers/tworooms/controller.pt` | TwoRooms controller β 100% |
|
| 103 |
| `checkpoints/base_world_models/pusht/weights.pt` | Frozen PushT world model (LeWM) |
|
|
|
|
| 25 |
|
| 26 |
| Environment | Method | Success | Notes |
|
| 27 |
|---|---|---|---|
|
| 28 |
+
| PushT | LePlanner (rh=1) | **98%** | arrival+hold checkpoint |
|
| 29 |
+
| PushT | LePlanner (rh=5) | **92%** | |
|
| 30 |
| PushT | CEM (rh=5) | 90% | baseline search planner |
|
| 31 |
| Reacher | LePlanner (rh=1) | **100%** | 50 episodes, seed 42 |
|
| 32 |
| TwoRooms | LePlanner (rh=1) | **100%** | 50 episodes, seed 42 |
|
|
|
|
| 97 |
|
| 98 |
| File | Description |
|
| 99 |
|---|---|
|
| 100 |
+
| `checkpoints/ablations/ah_hold0.5/controller.pt` | **Headline PushT controller** β 98% at rh=1, 92% at rh=5 |
|
| 101 |
| `checkpoints/environment_controllers/reacher/controller.pt` | Reacher controller β 100% |
|
| 102 |
| `checkpoints/environment_controllers/tworooms/controller.pt` | TwoRooms controller β 100% |
|
| 103 |
| `checkpoints/base_world_models/pusht/weights.pt` | Frozen PushT world model (LeWM) |
|