Tock 1
Current issues?
We have from now until ~January to clean up Bessy. What should we be doing?
Brian
- Was looking at refactoring EEG_data. First step might be to pull out the EEG and marker buffers out so that anything can be the input (ex: not just lsl). Ultimately this would allow some one else to work with Bessy without needing to send LSL messages.
- Make Bessy interoperable with MOABB. Users could train a classifier offline with MOABB and use inside Bessy.
Anup
- Agree with focus on what Bessy does best, ex: realtime use case.
- Concern over what we focus on first, ex: how much time do we spend on various items. Wants to have a plan that we’re working on together.
- Thinks there’s a way to be better than MOABB, ex: model hub / cartridge hub. It’s important to avoid being dependent on this. More discussion needed re: training data sets are typically small in the BCI world.
Eli
- Look at how inference is handled. Separate training from inference like some other tools do.
- Saving is a key problem for us right now.
- Making it easier for someone to understand how to use Bessy for a given task. We have documentation for functions now, but not much to connect the dots.
- Improving reliability of Bessy. If you deviate from a use case, it breaks. If you miss a packet it breaks.
- Provide a set of pre-defined scripts for creating their scene in Unity at runtime (not in the editor).
What is Bessy? What goes on the Box?
- A software suite for making object in Unity selectable with a BCI
- A suite of tools for quickly creating applications or games while abstracting out the BCI back end.
- A set of tools for realtime training of classifiers, perform inferences, manage data in realtime.
- A layer of abstraction that tailors BCI to our own use cases
- Works in offline / online use cases in an identical fashion
- Independence of front end and back end, thereby allowing someone to change the BCI elements and the game / application stays the same.