Learn
Explore
Toggle theme
Sign in
Create account
Back to Courses
Progress
Loading...
Back to Courses
Course
VLA Robotics Course
About this course
43 lessons
3 modules
1.
VLA Intro
Finetuning Vision-Language-Action (VLA) Models at Scale with Ray on Anyscale
What you are optimizing
How to interpret the scale of this tutorial
Import core libraries and Ray primitives
+1 more lesson
2.
DeepSeek
Finetuning Vision-Language-Action Models at Scale with DeepSpeed and Ray on Anyscale
What you are optimizing
How to interpret the scale of this tutorial
3.
BEV
Training Bird’s-Eye View (BEV) Models for Robotics at Scale with Ray on Anyscale
Cell 1: Define NuScenes storage paths and validate your runtime
Cell 2: Download large datasets reliably with resume support
Cell 3: Download the NuScenes mini dataset archive
Cell 4: Safely extract the NuScenes dataset into cluster storage
Cell 5: Initialize the NuScenes dataset and validate your installation
Cell 6: Inspect dataset scale and enumerate available scenes
Cell 7: Inspect a single scene sample and its sensor layout
Cell 8: Visualize camera annotations for a single timestep
Cell 9: Visualize lidar data in the ego-centric top-down frame
Cell 10: Project lidar points into the camera image plane
Cell 11: Visualize lidar intensity projected into the camera image
Cell 12: Render all camera views for a single timestep
Cell 13: Create a lightweight subset manifest for scalable training
Cell 14: Import core libraries for Ray Data + Ray Train BEV training
Cell 15: Define persistent storage paths and select your subset manifest
Cell 16: Define your BEV grid, camera inputs, and label space
Cell 17: Build a lightweight training manifest from NuScenes sample tokens
Cell 18: Split the manifest into training and validation sets
Cell 19: Preprocess NuScenes samples into fixed-shape BEV training tensors
Cell 20: Build Ray Data datasets for distributed preprocessing
Cell 21: Run a one-batch sanity check on your Ray Data pipeline
Cell 22: Define a minimal camera-only BEV Transformer and validate shapes
Cell 23: Define the Ray Train worker loop with DDP, mixed precision, and checkpoint resume
Cell 24: Launch distributed training with TorchTrainer and persist results
Conclusion
+23 more lessons
