Building a production-grade VLA system for household manipulation on the Franka Research 3 robot.
This project tracks the full pipeline: data collection, model training, sim-to-real transfer, and real-world deployment with measurable autonomy metrics.
Progress updates:
- Demos — continuous-shot videos of real robot deployment
- Metrics — autonomy %, success rate, cycle time
- Blog posts — technical deep-dives into methods and lessons
Demo
TODO
Things I Learned
TODO
Written Output
Researching robot learning led to these technical deep-dives:
- Why the Gymnasium API Looks the Way It Does?
- Why Naive Behavioral Cloning Doesn’t Work?
- A Beginner’s Guide to Flow Matching
- Understanding Action Chunking with Flow Matching
- How to Derive the Policy Gradient with Monte Carlo Sampling?