Commit
•
11b31cc
1
Parent(s):
b8a0728
Update README.md
Browse files
README.md
CHANGED
|
@@ -581,24 +581,24 @@ test_dataset = load_dataset("EpicPinkPenguin/procgen", name="bossfight", split="
|
|
| 581 |
## Agent Performance
|
| 582 |
The PPO RL agent was trained for 50M steps on each environment and obtained the following final performance metrics.
|
| 583 |
|
| 584 |
-
| Environment | Return |
|
| 585 |
-
|
| 586 |
-
| bigfish | 29.16 |
|
| 587 |
-
| bossfight | 11.35 |
|
| 588 |
-
| caveflyer | 09.47 |
|
| 589 |
-
| chaser | 11.46 |
|
| 590 |
-
| climber | 11.17 |
|
| 591 |
-
| coinrun | 09.74 |
|
| 592 |
-
| dodgeball | 16.78 |
|
| 593 |
-
| fruitbot | 29.87 |
|
| 594 |
-
| heist | 09.98 |
|
| 595 |
-
| jumper | 08.71 |
|
| 596 |
-
| leaper | 07.71 |
|
| 597 |
-
| maze | 09.99 |
|
| 598 |
-
| miner | 12.63 |
|
| 599 |
-
| ninja | 09.44 |
|
| 600 |
-
| plunder | 25.98 |
|
| 601 |
-
| starpilot | 55.28 |
|
| 602 |
|
| 603 |
|
| 604 |
## Dataset Structure
|
|
@@ -651,27 +651,5 @@ The dataset is divided into a `train` (90%) and `test` (10%) split. Each environ
|
|
| 651 |
## Dataset Creation
|
| 652 |
The dataset was created by training an RL agent with [PPO](https://arxiv.org/abs/1707.06347) for 50M steps in each environment. The trajectories where generated by sampling from the predicted action distribution at each step (not taking the argmax). The environments were created on `distribution_mode=easy` and with unlimited levels.
|
| 653 |
|
| 654 |
-
## Video Samples
|
| 655 |
-
Here is a collection of videos with the RGB observations from the dataset.
|
| 656 |
-
|
| 657 |
-
| Environment | Observation |
|
| 658 |
-
|:------------|:------------|
|
| 659 |
-
| bigfish | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/lHQXBqLdoWicXlt68I9QX.mp4"></video> |
|
| 660 |
-
| bossfight | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/LPoafGi4YBWqqkuFlEN_l.mp4"></video> |
|
| 661 |
-
| caveflyer | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/XVqRwu_9yfX4ECQc4At4G.mp4"></video> |
|
| 662 |
-
| chaser | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/FIKVv48SThqiC1Z2PYQ7U.mp4"></video> |
|
| 663 |
-
| climber | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/XJQlA7IyF9_gwUiw-FkND.mp4"></video> |
|
| 664 |
-
| coinrun | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/Ucv3HZttewMRQzTL8r_Tw.mp4"></video> |
|
| 665 |
-
| dodgeball | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/5HetbKuXBpO-v1jcVyLTU.mp4"></video> |
|
| 666 |
-
| fruitbot | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/zKCyxXvauXjUac-5kEAWz.mp4"></video> |
|
| 667 |
-
| heist | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/AdZ6XNmUN5_00BKd9BN8R.mp4"></video> |
|
| 668 |
-
| jumper | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/s5k31gWK2Vc6Lp6QVzQXA.mp4"></video> |
|
| 669 |
-
| leaper | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/_hDMocxjmzutc0t5FfoTX.mp4"></video> |
|
| 670 |
-
| maze | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/uhNdDPuNhZpxVns91Ba-9.mp4"></video> |
|
| 671 |
-
| miner | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/ElpJ8l2WHJGrprZ3-giHU.mp4"></video> |
|
| 672 |
-
| ninja | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/b9i-fb2Twh8XmBBNf2DRG.mp4"></video> |
|
| 673 |
-
| plunder | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/JPeGNOVzrotuYUjfzZj40.mp4"></video> |
|
| 674 |
-
| starpilot | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/wY9lZgkw5tor19hCWmm6A.mp4"></video> |
|
| 675 |
-
|
| 676 |
## Procgen Benchmark
|
| 677 |
The [Procgen Benchmark](https://openai.com/index/procgen-benchmark/), released by OpenAI, consists of 16 procedurally-generated environments designed to measure how quickly reinforcement learning (RL) agents learn generalizable skills. It emphasizes experimental convenience, high diversity within and across environments, and is ideal for evaluating both sample efficiency and generalization. The benchmark allows for distinct training and test sets in each environment, making it a standard research platform for the OpenAI RL team. It aims to address the need for more diverse RL benchmarks compared to complex environments like Dota and StarCraft.
|
|
|
|
| 581 |
## Agent Performance
|
| 582 |
The PPO RL agent was trained for 50M steps on each environment and obtained the following final performance metrics.
|
| 583 |
|
| 584 |
+
| Environment | Return | Observation |
|
| 585 |
+
|:------------|:-------|:------------|
|
| 586 |
+
| bigfish | 29.16 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/lHQXBqLdoWicXlt68I9QX.mp4"></video> |
|
| 587 |
+
| bossfight | 11.35 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/LPoafGi4YBWqqkuFlEN_l.mp4"></video> |
|
| 588 |
+
| caveflyer | 09.47 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/XVqRwu_9yfX4ECQc4At4G.mp4"></video> |
|
| 589 |
+
| chaser | 11.46 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/FIKVv48SThqiC1Z2PYQ7U.mp4"></video> |
|
| 590 |
+
| climber | 11.17 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/XJQlA7IyF9_gwUiw-FkND.mp4"></video> |
|
| 591 |
+
| coinrun | 09.74 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/Ucv3HZttewMRQzTL8r_Tw.mp4"></video> |
|
| 592 |
+
| dodgeball | 16.78 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/5HetbKuXBpO-v1jcVyLTU.mp4"></video> |
|
| 593 |
+
| fruitbot | 29.87 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/zKCyxXvauXjUac-5kEAWz.mp4"></video> |
|
| 594 |
+
| heist | 09.98 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/AdZ6XNmUN5_00BKd9BN8R.mp4"></video> |
|
| 595 |
+
| jumper | 08.71 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/s5k31gWK2Vc6Lp6QVzQXA.mp4"></video> |
|
| 596 |
+
| leaper | 07.71 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/_hDMocxjmzutc0t5FfoTX.mp4"></video> |
|
| 597 |
+
| maze | 09.99 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/uhNdDPuNhZpxVns91Ba-9.mp4"></video> |
|
| 598 |
+
| miner | 12.63 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/ElpJ8l2WHJGrprZ3-giHU.mp4"></video> |
|
| 599 |
+
| ninja | 09.44 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/b9i-fb2Twh8XmBBNf2DRG.mp4"></video> |
|
| 600 |
+
| plunder | 25.98 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/JPeGNOVzrotuYUjfzZj40.mp4"></video> |
|
| 601 |
+
| starpilot | 55.28 | <video controls autoplay loop src="https://cdn-uploads.huggingface.co/production/uploads/633c1daf31c06121a58f2df9/wY9lZgkw5tor19hCWmm6A.mp4"></video> |
|
| 602 |
|
| 603 |
|
| 604 |
## Dataset Structure
|
|
|
|
| 651 |
## Dataset Creation
|
| 652 |
The dataset was created by training an RL agent with [PPO](https://arxiv.org/abs/1707.06347) for 50M steps in each environment. The trajectories where generated by sampling from the predicted action distribution at each step (not taking the argmax). The environments were created on `distribution_mode=easy` and with unlimited levels.
|
| 653 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 654 |
## Procgen Benchmark
|
| 655 |
The [Procgen Benchmark](https://openai.com/index/procgen-benchmark/), released by OpenAI, consists of 16 procedurally-generated environments designed to measure how quickly reinforcement learning (RL) agents learn generalizable skills. It emphasizes experimental convenience, high diversity within and across environments, and is ideal for evaluating both sample efficiency and generalization. The benchmark allows for distinct training and test sets in each environment, making it a standard research platform for the OpenAI RL team. It aims to address the need for more diverse RL benchmarks compared to complex environments like Dota and StarCraft.
|