HKUST PhD Chronicle, Week 48, Drowning in Experiments

Experiments

This week I was drowning in experiments. I have already started documenting my experiments in a git-tracked repo, but maintaining them takes quite a lot of time. It reminds me of what Shunyu Yao stated in a recent podcast:

Quote

The current bottleneck in AI development lies not in a lack of ideas, but in having too many ideas that each require substantial time to validate through experiments, one by one.

I enjoy doing experiments, but having tons of them to track is making me overwhelmed.

ROS2 Network

From time to time, I have been haunted by the communication_constraints_violation error in libfranka: C++ library for Franka Robotics research robots:

const bool & 	communication_constraints_violation

This happens because of the network topology in our lab when using remote inference. Recently, I found this post, which tremendously reduces the chances of this error occurring.

Realtime Action Chunking

Finally, it seems I have almost cracked the algorithm in the 📄Real-Time Execution of Action Chunking Flow Policies from the bottom up. I will release a detailed post about it soon!