Abstract An all-chain-wireless brain-to-brain system (BTBS), which enabled motion control of a cyborg cockroach via human brain, was developed in this work. Steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI) was used in this system for recognizing human motion intention and an optimization algorithm was proposed in SSVEP to improve online performance of the BCI. The cyborg cockroach was developed by surgically integrating a portable microstimulator that could generate invasive electrical nerve stimulation. Through Bluetooth communication, specific electrical pulse trains could be triggered from the microstimulator by BCI commands and were sent through the antenna nerve to stimulate the brain of cockroach. Serial experiments were designed and conducted to test overall performance of the BTBS with six human subjects and three cockroaches. The experimental results showed that the online classification accuracy of three-mode BCI increased from 72.86% to 78.56% by 5.70% using the optimization algorithm and the mean response accuracy of the cyborgs using this system reached 89.5%. Moreover, the results also showed that the cyborg could be navigated by the human brain to complete walking along an S-shape track with the success rate of about 20%, suggesting the proposed BTBS established a feasible functional information transfer pathway from the human brain to the cockroach brain.

Citation: Li G, Zhang D (2016) Brain-Computer Interface Controlled Cyborg: Establishing a Functional Information Transfer Pathway from Human Brain to Cockroach Brain. PLoS ONE 11(3): e0150667. https://doi.org/10.1371/journal.pone.0150667 Editor: Jacob Engelmann, Universität Bielefeld, GERMANY Received: September 4, 2015; Accepted: February 16, 2016; Published: March 16, 2016 Copyright: © 2016 Li, Zhang. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: All relevant data are within the paper. Funding: This work was supported by the National Natural Science Foundation of China (No. 41751475292), the National Basic Research Program (973 program) of China (2011CB013305), and the National High Technology Research and Development Program (863 Program) of 419 China (2015AA020501). Competing interests: The authors have declared that no competing interests exist.

Introduction The movie “Avatar” shows the dream of direct brain-to-brain control between different individuals, which has inspired researchers to develop physical communication connections between different brains. Some pilot studies have moved a step forward toward this fantastic dream through developing a brain-to-brain system (BTBS). Pais-Vieira et al. [1] developed a BTBS between two rats based on invasive technologies (neuronal ensemble recordings and intracortical microstimulation), where the “decoder” rat could follow the same sensorimotor task as the “encoder” rat with the system. Yoo et al. [2] established functional interface between brains of human and rat based on noninvasive technologies, electroencephalography (EEG) and focused ultrasound (FUS), where the rat’s tail could be controlled by a human’s intention. Rao et al. [3] realized a BTBS between two persons based on noninvasive technologies, EEG and transcranial magnetic stimulation (TMS), in which the receiver’s (a subject) finger motion could be triggered by a sender (the other subject). The previous studies are amazing, but the function of such BTBS is very limited, i.e. only a simple motion or action can be accomplished. This work aims to develop a versatile multi-functional BTBS between humans and cockroaches. A typical BTBS should consist of two key parts: brain-computer interfaces (BCIs) used for getting information from one brain and neuromodulation technologies used for sending information to another brain. BCI is a well-known technology that helps people communicate with external worlds through measuring and translating brain signals without involving muscular movements or peripheral nervous system [4]. Using BCIs, humans have already could control external devices such as robotic arm, wheel chair, quadcopter and so on [5–7]. In general, there are two categories of BCIs, i.e. invasive and non-invasive BCIs. Non-invasive BCIs include EEG, near-infrared spectroscopy (NIRS), and functional magnetic resonance imaging (fMRI) etc. While, electrocorticography (ECoG) and intracortical recordings belong to the scope of invasive BCIs [8, 9]. Among the available BCIs, EEG should be the most practical and convenient method at present, and steady-state visual evoked potential (SSVEP) is one of the most reliable, versatile and robust EEG paradigms, which is thus adopted in our study [10]. SSVEP is an EEG response from the visual cortex generated by repetitive visual stimuli when subjects gaze at flickering source [11]. SSVEP has high signal-to-noise ratio (SNR) and information transfer rate (ITR), and needs short training time. These favorable features of SSVEP are desired for the high-performance BTBS developed in this study. Even though SSVEP can be elicited by visual stimulation with a wide range of frequencies from 1–100 Hz, the better SSVEP performance is achieved under visual stimulation with lower frequencies range of 6–15 Hz in general [12, 13]. Neuromodulation, as an intervention technology used to modulate the nervous system function, has been applied to the brain, spinal cord, peripheral nerves, autonomic nerves and other nervous organs [14]. Among the variety of technologies used for neuromodulation (mechanically, chemically and electrically, etc.), electrical stimulation-based neuromodulation is frequently introduced when developing live cyborgs or biobots. In fact, a variety of cyborgs (a cyborg is an organism with both biological and electronic parts) have been successfully developed via applying electrical stimulation to specific muscles or nerves in recent years, such as rats, moths, and beetles [15–19]. To build up a cyborg that could be remotely controlled by a human brain, cockroach is a preferable choice for robust control performance and good surgical implementability. It’s known that cockroaches rely on antennas heavily for navigation during walking. The antenna is a multi-functional and important sensory organ that can generate tactile, thermal, humidity and olfactory sense. When sending specific micro-electrical pulse trains through the antenna nerve, stimulation information will be encoded as sensory information by the antennal neurons and sent to the brain, which will activate the descending mechanosensory interneurons (DMIs) (interneurons with the largest caliber axons descending to thoracic levels from the brain) [20, 21]. Subsequently, the thoracic motor centers will be activated, and as a result, the evasive behaviors such as turning will be elicited [21–24]. If an electrical pulse train is applied to the nerves of right (left) antenna, a left (right) -direction turn will be triggered from the cockroach. By inserting a tiny electrode into antenna and providing electrical stimulation, a functional neural-machine interface between external device and cockroach brain can be developed. Based on the present technologies of BCI and neuromulation, we find it is possible to realize communication and virtual signal transfer from one brain to another one with a functional BTBS. This interesting concept is explored in this study and we aim to control cyborg cockroaches to accomplish path-tracking tasks with human brain signals for the first time.

Discussion The present study demonstrated that human’s intention, extracted from the subject’s EEG, could be transmitted to partially dominate an insect’s brain. The human could successfully steer a cockroach to make turns using thoughts through SSVEP-based BCI and micro invasive neural stimulation. The SSVEP could achieve a high CA (90.21%) by only using 4 channels on the scalp of visual cortex with a portable and wireless EEG device in the training phase (Fig 5). We also proposed an optimization algorithm for SSVEP, which greatly eliminated the random errors occurred during online control process, as a result, it significantly increased online classification accuracy by 8.08% from 83.34% to 91.42% with the removal of frequency-transient state signals. Generally, misclassification appeared during shifting period of a subject’s gaze in SSVEP. Therefore this phenomenon might be improved by adding a program to detect transition of the EEG signal on frequency before classification in further study, even though such increase of online classification accuracy was possibly at the sacrifice of SSVEP program’s fast-response performance. Besides, regarding the present SSVEP-based BCI, due to high brightness and repetitive stimulation, the user might suffer fatigue when using SSVEP for a long time, where fatigue was a generally feeling of tiredness, reduced concentration when performing a task [31]. The amplitude and SNR of the EEG signals were consequently affected by the fatigue [32, 33]. One possible solution to reduce fatigue of SSVEP was to use visual stimulation with high frequency or modulated amplitude [34, 35]. By implementing the possible improvements on SSVEP in further research, we can move current system a step closer to practical applications. The other similar studies were compared with our work as well. Pais-Vieira et al. [1] succeeded in building up a BTBS for real-time sharing of sensorimotor information between two rats with intra-cortical microstimulation (ICMS), in which the decoder translated correctly 64.32% the motor information (two-mode) from the encoder under a time delay of 20.06 s (encoder) and 13.59 s (decoder), another experiment in the research demonstrated the decoder translated correctly encoder’s 62.34% tactile information (two-mode) and the time delay was 2.66 s (encoder) and 2.68 s (decoder) respectively. Recently, Rao et al. [3] accomplished a direct BTBS between humans, where the motor imaginary (two-mode) and TMS were applied to complete a task, the paired true positive rate (TPR) varied from 83.33% to 25.00% and paired false positive rate (FPR) varied from 0.00% to 37.50% in online experiments with a 2 s set-up time plus a 20 s visual countdown in each trial and a transmission duration of about 650 ms along the system. In this work, cyborgs (receiver) produced quick and accurate response (mean RA = 89.5%, left-turn accuracy (84.6%), right-turn accuracy (92.0%)) through applying the invasive technique of neuromodulation, where monopole square pulse trains were indirectly sent to the brains of the cyborg cockroach via antennas, and the time that a command sent from SSVEP to the completion of reaction from the cyborg was about 772 ms. In addition, the techniques we used in BCI (three-mode) segment in this study showed high decoding accuracy (CA = 86.0±10.4%) of sender’s intention as well and the average decoding time was 1.79 s (0.25 s minimum). In the work of [22], Latif et al. tried to steer a cyborg cockroach to walk along an S-shape lap in two-directions manually and succeeded in achieving a success rate of around 10%. However, a successful online navigation of cyborg with human brain in this study was much more challenging, which required stable and continuous high level of accuracy in both “sender” and “receiver” sides, the experimental results demonstrated the BCI could successfully control the cyborg to walk along the S-shape track in all three groups, and the success rate ranged from 0.00% to 50.00% in 10-trial experiments, where the mean success rate (SR) was 20.00%, the value was not high enough, but significantly above the level in control group (0.00%) (t = 3.464, P = 0.0085). And the SR coincided with the roughly estimated value ((CA*RA)TC) based on the accuracy from both “sender” and “receiver” parts as there were 12.7 turning commands in average (Table 2) during each online trial. Additionally, the CPC, as an indicator of system performance, reached 0.616 (Table 2), while its chance level was 0.375. Above all, the results achieved through the system in present experiments were rather optimistic and instructive. In addition, the architecture of current system still had the potential of enabling developing multi-mode control cyborgs (more than three modes), and increasing the control modes might contribute to the improvement of SR in online control tasks and at the same time enable more precise online assignments as well, which would be designed and validated in further study. It is meaningful to build up a BTBS between the human and the animal. Firstly, this all-chain-wireless system proves that it’s possible to built up an embryonic virtual brain-to-brain information transfer device with current technology. Besides, the idea also has the applicable value in reality, e.g. could be further used for detection in complex and dangerous environments. Most importantly, as a pioneer study of brain communication between human and animal, the successful implementation of BTBS expands the possible usage of both BCI and neuromodulation, furthermore, the alternative BCI and neuromodulation modalities (both invasive and non-invasive) may have huge potential of intersection and integration to create far-reaching usage in future. We have seen a mushroom growth of internet based on computers during recent years, and perhaps brain-to-brain network may be developed in the future just as the science-fiction films predict.

Acknowledgments This work was supported by the National Natural Science Foundation of China (No. 51475292), the National Basic Research Program (973 program) of China (2011CB013305), the National High Technology Research and Development Program (863 Program) of China (2015AA020501). The authors would like to thank all the subjects for their participation in the experiments.

Author Contributions Conceived and designed the experiments: DGZ. Performed the experiments: GYL. Analyzed the data: GYL. Contributed reagents/materials/analysis tools: DGZ GYL. Wrote the paper: DGZ GYL.