Key points are not available for this paper at this time.
Although brain–computer interfaces (BCIs) have made remarkable progress in recent decades, there remains no universal BCI framework that supports efficient translation across diverse neural signals and decoding algorithms from offline decoding to real‐time online control, while also enabling the integration of neural mechanisms. To address these challenges, this work presents ControlIt, a universal BCI framework that bridges laboratory research and clinical applications, supports both fixed and adaptive decoders, and enables seamless transition from offline training to online neural adaptation. ControlIt is implemented using Robot Operating System 2 (ROS2) and consists of three modular components—Observation, State, and Decoder—each running as an independent ROS2 node that can be flexibly combined. For spike‐based regression BCIs, it supports both traditional decoding methods and deep learning approaches with real‐time model adaptation. For electroencephalography (EEG)‐ and electrocorticography (ECoG)‐based classification BCIs, ControlIt enables single‐band and multiband online decoding with interblock adaptation. Importantly, ControlIt maintains low and stable communication and inference latency across all signal modalities. This versatile architecture provides a robust foundation for future comprehensive BCI studies and applications.
Yang et al. (Tue,) studied this question.