Geoff Lord has embarked on an engineering adventure that involved turning a children’s toy into an EEG, training neural networks on the toy’s extracted data, and a successful attempt at mapping less-than-ideal neural activity data to intended actions.
I thought back to a toy my brother and I used to play with when we were young, the Mattel Mindflex. The Mindflex was a game where the users would don headsets that included an EEG capable of reading neural activity. This neural activity could then be used to manipulate a ball around a play area. After a brief search, it became evident that others had hacked apart these headsets to expose the EEG data they generated.
The Mindeflex headset required a slight hardware modification to gain access to the EEG sensor data. First, the device was disassembled which exposed the Neurosky EEG chip. Subsequently, this allowed for a small jumper wire to be soldered onto the chip’s TX pin. This TX pin was then connected to a microcontroller that takes in the EEG byte stream from the Neurosky chip and outputs parsed ASCII text to a console.
The parsing of the data was made easy by a preexisting library called Brain. The microcontroller was outfitted with a push button to allow for data collection analogous to the data used in the Neuralink monkey pong demonstration. While this experimental apparatus is not ideal, it provided a way to generate datasets at no cost.
Check out more in the post here.