Stanford researchers improve brain machine interface's versatility, speed, and accuracy to previously unreported levels. The novel algorithm uses a continuous decoder and a discrete hidden markov model (HMM) state detector. The system can infer many states such as a click state, an idle/go/stop state or a state that manipulates a decoder's parameters or velocity. Additionally, the system can optimally tune the decoder's parameters, e.g., its states and transition probabilities, to better suit the subject. This invention is a promising alternative to hold-time based decoders.
Technology developed by Stanford researchers directly reads brain signals to drive a cursor moving over a keyboard. Figure:
More complex state models are possible, with different states. For example, we have also decoded an "idle" state, a "move slow" state, and a "move fast" state. These are useful, e.g. when decoding that the subject is idling, and does not want to do the task, we can effectively "turn off" the task.
Stage of Research
-Successful closed-loop implementation in 2 live subjects
- Throughput performance increases of up to 50% in monkeys
- Translation to human participant, demonstrated useful performance benefit, pending publication