Recent advancements in brain-computer interface (BCI) technology have marked a significant milestone in real-world applications, with researchers from the Chinese Academy of Sciences’ Center for Excellence in Brain Science and Intelligent Technology successfully completing the second clinical trial using an invasive BCI system. The team employed the high-throughput wireless invasive BCI system, WRS01, enabling a patient with high-level paralysis—caused by spinal cord injury in 2022—to independently control a smart wheelchair and a robotic dog using only brain signals. The implantation of the BCI system took place in June 2025, and after several weeks of training, the patient achieved stable control of a computer cursor and a tablet interface. Building on this progress, the research team extended the system’s capabilities to three-dimensional physical devices, achieving continuous, stable, and low-latency control of both the smart wheelchair and the robotic dog. This allowed the patient to perform various functional tasks in complex real-life environments, demonstrating the system’s practical utility in daily living. The study achieved multiple technical breakthroughs. In neural signal decoding, the team developed a high-compression, high-fidelity neural data compression method, integrating spike-band power, inter-spike interval, and spike count features into a hybrid decoding model. This approach maintains high performance even under noisy conditions, improving overall brain control accuracy by 15% to 20%. To address signal instability caused by environmental interference—such as sound, light, electromagnetic noise—and fluctuations in the patient’s physiological and psychological states, the team introduced neural manifold alignment. This technique extracts stable low-dimensional features from high-dimensional, dynamic neural signals, significantly enhancing the decoder’s adaptability and stability across days. The team also developed an online re-calibration system that allows real-time adjustment of decoding parameters during daily use, eliminating the need for dedicated calibration sessions and enabling a seamless, “ever-improving” user experience—where performance gets better with continued use. Through a custom communication protocol, the system reduced end-to-end latency from signal acquisition to command execution to under 100 milliseconds—below the threshold of human perceptual delay—resulting in a highly responsive and natural control experience. Neuroscientific analysis revealed that as the patient gained proficiency, task-related neural activity shifted from widespread neuronal engagement to a focused pattern involving only a few efficient neurons. This transition reduced cognitive load and enabled the patient to “internalize” control of the external devices, providing a neural basis for the perception of “thought-driven” action. In terms of real-world integration, the team collaborated with local disability associations to enable the patient to participate in online data annotation tasks, promoting social reintegration. The team is also advancing the technology’s industrialization through a systematic approach, starting with neural interface electrodes and building a full technology stack encompassing system integration, algorithm optimization, and application development. Based on the data and experience from the current trial, the team has developed an upgraded system, WRS02, which increases the number of recording channels to 256. The first clinical trial for WRS02 is scheduled to begin shortly. With ongoing clinical data collection, high-quality neural-behavioral datasets will continue to drive algorithm improvements and the development of new application scenarios, establishing a self-reinforcing cycle of data-driven innovation. The system’s development is supported by micro- and nano-electronic fabrication platforms for producing flexible, implantable BCI electrodes.