Update README.md

README.md

@@ -32,7 +32,35 @@ The pre-trained RDT model can be fine-tuned for specific robotic embodiment and
 Here's an example of how to use the RDT-1B model for inference on a Mobile-ALOHA robot:
 
 ```python
-# Clone the repo and install depencies
+# Clone the repository and install dependencies
+from scripts.agilex_model import create_model
+CAMERA_NAMES = ['cam_high', 'cam_right_wrist', 'cam_left_wrist'] # Names of cameras used for visual input
+config = {
+    'episode_len': 1000,  # Length of one episode
+    'state_dim': 14,      # Dimension of the robot's state
+    'chunk_size': 64,     # Number of actions to predict in one step
+    'camera_names': CAMERA_NAMES,
+}
+ctrl_freq=25 # Set the control frequency (Hz)
+pretrained_vision_encoder_name_or_path = "google/siglip-so400m-patch14-384" # The pre-trained vision encoder model
+# Create the model with specified configuration
+model = create_model(
+    args=config,
+    dtype=torch.bfloat16,  # Use bfloat16 for improved performance
+    pretrained_vision_encoder_name_or_path=pretrained_vision_encoder_name_or_path,
+    control_frequency=ctrl_freq,
+)
+# Start inference process
+lang_embeddings_path = 'your/language/embedding/path' 
+text_embedding = torch.load(lang_embeddings_path)['embeddings'] # Load pre-computed language embeddings
+images: List(PIL.Image) = ... #  The images from last 2 frame
+proprio = ... # The current robot state
+# Perform inference to predict the next chunk_size actions
+actions = policy.step(
+    proprio=proprio,
+    images=images,
+    text_embeds=lang_embeddings 
+)
 ```
 
 RDT-1B supports finetuning on custom dataset, deploying and inferencing on real-robots, as well as pretraining the model.