DeepBrain: Enabling Fine-Grained Brain-Robot Interaction through Human-Centered Learning of Coarse EEG Signals from Low-Cost Devices

With the recent advancements of electroencephalograph (EEG) techniques, some brain-computer interface (BCI) solutions have been explored to assist individuals performing various tasks with their minds. One promising application is to combine BCI with robotic systems so that the mobility-impaired people can control robots to take care of themselves. Towards this ultimate goal to design BCIs for mobility-impaired, we firstly conducted an online survey with 54 mobility-impaired participants who barely had previous experience with BCI to identify the challenges they face in life for the purpose of designing a personalized BCI system in need. The results revealed these challenges including small daily tasks (such as feeding and cleaning), which weigh on the financial burdens of hiring a caregiver. Meanwhile, the off-the-shelf high-fidelity BCIs are often expensive, whereas the cheaper devices only collect coarse-grained signals, preventing practical application in care aids due to lack of temporal resolution and accuracy. Based on the survey findings, we then designed DeepBrain, a human-centered learning augmented BCI system, that requires only coarse-grained brain signals with low-cost BCI equipment, but supports fine-grained brain-robot interaction and scalable multi-robot collaboration for domestic multi-task operations. A follow-up system comparison with other approaches show that the proposed human-centered solution is a promising step towards the ultimate goal, as it achieves satisfactory accuracy with less low computation resources. Also the practical brain to multi-robot interaction system validates the feasibility of our framework and model used in DeepBrain.