Tips on Mental Command training
The way to get the best out of mental commands is to follow the instructions and practice the kind of things that work best for you – the new training mode in EmotivBCI provides some feedback about how distinguishable each command is from the others.
We deliberately do not restrict the kind of signals that can be used for each detection. I’ve consulted with our research team and here are the most important things:
1. Each action produces a distinct pattern from every other action. This is visible in the positioning of the dots in the EmotivBCI training screen (similar detections are closer together on the map). We find the best way to do this is to use very different images for each action. For example, think of the taste of lemons for PUSH and putting your hand in hot water for PULL. Many people get good results just by imagining physical motions of something, such as a ball of fluid mounted in the middle of your brain that you try to squeeze through holes in each direction, etc.
2. The visualizations should be easy to reproduce each time. This comes with the practice generally. We find people can make one or two actions pretty easily with moderate accuracy but with practice (and perhaps by making a brand new profile after you get the hang of it) you can learn to make repeatable, distinguishable mental commands. This can take several hours of practice to become an expert, but even 20 minutes of practice is enough to get a pretty good result.
We use a whole bunch of different features in the dataset (e.g power in a lot of narrow frequency bands, for each sensor, and several other kinds of feature) and then we clustering algorithms to predict the probability cloud of state, which is compared to the real-time data when you try to use the detection.