update

app.py

@@ -0,0 +1,77 @@
+import gradio as gr
+from benchmarks import benchmarks, update_environments
+from model import submit_model
+
+def clear_form():
+    return [None, None, None, None, None, None]
+
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("<center><h1 style='font-size: 40px;'>⚔️Reinforcement Learning Agent Arena⚔️</h1></center>")
+    gr.Markdown("""
+                # 📜📜📜 The workflow of RLArena
+                - Select a benchmark, an environment, and a specific version.
+                - Enter your github username and the link to your code.
+                - Click "Submit" to evaluate your agent.
+                # 🏆🏆🏆 Visit the leaderboard
+                - Accelerate your RL research with the well-organized benchmark scores.
+                # 💪💪💪 Submit your agent now!
+                """)
+    with gr.Row():
+        with gr.Column(scale=2):
+            with gr.Row():
+                github_username = gr.Textbox(label="Github Username", info="Please enter your github username, e.g., username.")
+                benchmark = gr.Dropdown(label="Benchmark", choices=list(benchmarks.keys()), info="Please select a benchmark, e.g., Procgen.")
+            with gr.Row():
+                environment = gr.Dropdown(label="Environment", choices=[], value=None, info="Please select an environment, e.g., Miner.")
+                version = gr.Dropdown(label="Version", choices=[], value=None, info="Please select a version, e.g., v0.")
+            with gr.Row():
+                training_steps = gr.Number(label="Training Steps", precision=0, info="Please enter the training steps, e.g., 1000000.")
+                code_link = gr.Textbox(label="Code Link", info="Example: https://github.com/username/repo, the link should be accessible.")
+            with gr.Row():
+                submit_button = gr.Button("Submit", variant="primary") 
+                clear_button = gr.Button("Clear", variant="secondary")
+        with gr.Column(scale=1):
+            # file uploader
+            model_uploader = gr.File(label="Upload the agent here!")
+            output = gr.Textbox(label="Evaluation Result")
+    
+    benchmark.change(
+        fn=update_environments,  
+        inputs=benchmark,
+        outputs=[environment, version]
+    )
+
+    submit_button.click(submit_model, inputs=[github_username, benchmark, environment, version, training_steps, code_link, model_uploader], outputs=output)
+    clear_button.click(clear_form, inputs=[], outputs=[github_username, benchmark, environment, version, training_steps, code_link])
+
+    with gr.Row():
+        # add multiple images with html
+        html_images = """
+        <div><br><br></div>
+
+        # 🔥🔥🔥 Powered by
+        <div style="display: flex; flex-wrap: wrap; gap: 10px; justify-content: left;">
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_polyu.png" alt="Image 1" style="max-width: 100%; height: auto;">
+          </div>
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_sjtu.png" alt="Image 2" style="max-width: 100%; height: auto;">
+          </div>
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_eias.png" alt="Image 3" style="max-width: 100%; height: auto;">
+          </div>
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_idt.png" alt="Image 4" style="max-width: 100%; height: auto;">
+          </div>
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_ustc.png" alt="Image 5" style="max-width: 100%; height: auto;">
+          </div>
+          <div style="flex: 0 0 calc(25% - 10px); display: flex; justify-content: center; align-items: center;">
+            <img src="/file=static/logo_purdue.png" alt="Image 6" style="max-width: 100%; height: auto;">
+          </div>
+        </div>
+        """
+        gr.Markdown(html_images)
+        
+        
+demo.launch(allowed_paths=["./"])

benchmarks.py

@@ -0,0 +1,33 @@
+import gradio as gr
+import gymnasium as gym
+
+benchmarks = {
+    "Arcade Learning Environment": {
+        'env': ['Breakout', 'Pong', 'Qbert', 'Seaquest', 'SpaceInvaders'],
+        'version': ['NoFrameskip-v4'],
+    },
+    "Procgen": {
+        'env': ['BigFish', 'BossFight', 'Chaser', 'Climber', 'CaveFlyer', 'CoinRun', 'Dodgeball', 'FruitBot',  
+                'Heist', 'Jumper', 'Leaper', 'Maze', 'Miner', 'Ninja', 'Plunder', 'Starpilot'],
+        'version': ['Easy', 'Hard'],
+    },
+    'DeepMind Control Suite': {
+        'env': ['Humanoid_Stand', 'Humanoid_Walk', 'Humanoid_Run'],
+        'version': ['State-based', 'Image-based'],
+    },
+}
+
+def update_environments(benchmark):
+    if benchmark in benchmarks:
+        print(f"Updating environments for {benchmark}: {benchmarks[benchmark]}")
+        return gr.update(choices=benchmarks[benchmark]['env'], value=None), gr.update(choices=benchmarks[benchmark]['version'], value=None)
+    else:
+        print("No environments found for the selected benchmark")
+        return gr.update(choices=[], value=None), gr.update(choices=[], value=None)
+
+def build_env(benchmark, environment, version):
+    # if benchmark == "Arcade Learning Environment":
+    #     env = gym.make(f"{environment}-{version}")
+    # elif benchmark == "Procgen":
+    #     env = gym.make(f"{environment}-{version}")
+    return gym.make("CartPole-v1")

model.py

@@ -0,0 +1,61 @@
+import gradio as gr
+# import torch as th
+import numpy as np
+from benchmarks import build_env
+
+def submit_model(github_username, 
+                 benchmark, 
+                 environment, 
+                 version, 
+                 training_steps, 
+                 code_link,
+                 model_uploader
+                 ):
+    avg_episode_rewards = 0
+    success_msg = f"""
+    INFO: Submitted by {github_username}:
+    INFO: Benchmark: {benchmark}
+    INFO: Environment: {environment}
+    INFO: Version: {version}
+    INFO: Training Steps: {training_steps}
+    INFO: Code Link: {code_link}
+    INFO: Final Score: {avg_episode_rewards}
+    """
+    username_error_msg = f"""
+    ERROR: The GitHub username should be consistent with the code link!
+    """
+
+    model_none_error_msg = f"""
+    ERROR: No model uploaded!
+    """
+
+    # check if username is valid
+    if github_username.lower() not in code_link.lower():
+        return username_error_msg
+    if model_uploader is None:
+        return model_none_error_msg
+    
+    episode_rewards = evaluate_model(model_uploader, benchmark, environment, version)
+    avg_episode_rewards = np.mean(episode_rewards)
+
+    return success_msg
+
+def evaluate_model(model_uploader, benchmark, environment, version):
+    env = build_env(benchmark, environment, version)
+    print(env)
+
+    episode_rewards = []
+
+    obs, info = env.reset()
+    while len(episode_rewards) < 100:
+        # action = model_uploader.predict(obs)
+        action = env.action_space.sample()
+        obs, reward, terminated, truncated, info = env.step(action)
+        print(reward)
+        episode_rewards.append(reward)
+
+        if terminated or truncated:
+            obs, info = env.reset()
+
+    return episode_rewards
+

requirements.txt

@@ -0,0 +1,4 @@
+torch
+gymnasium
+numpy
+gradio