Abstract
We examine bidirectional brain-machine interfaces that control external devices in a closed loop by decoding motor cortical activity to command the device and by encoding the state of the device by delivering electrical stimuli to sensory areas. Although it is possible to design this artificial sensory-motor interaction while maintaining two independent channels of communication, here we propose a rule that closes the loop between flows of sensory and motor information in a way that approximates a desired dynamical policy expressed as a field of forces acting upon the controlled external device. We previously developed a first implementation of this approach based on linear decoding of neural activity recorded from the motor cortex into a set of forces (a force field) applied to a point mass, and on encoding of position of the point mass into patterns of electrical stimuli delivered to somatosensory areas. However, this previous algorithm had the limitation that it only worked in situations when the position-to-force map to be implemented is invertible. Here we overcome this limitation by developing a new non-linear form of the bidirectional interface that can approximate a virtually unlimited family of continuous fields. The new algorithm bases both the encoding of position information and the decoding of motor cortical activity on an explicit map between spike trains and the state space of the device computed with Multi-Dimensional-Scaling. We present a detailed computational analysis of the performance of the interface and a validation of its robustness by using synthetic neural responses in a simulated sensory-motor loop.
Introduction
Traditional Brain-Machine Interfaces (BMIs) typically operate as open loops, either decoding motor intent or encoding sensory feedback. Bidirectional BMIs (bBMIs) aim to close this loop, creating a more natural interaction between the brain and external devices. However, early bBMI implementations were constrained by linearity; they required an invertible relationship between a device's position and the force applied to it. This paper introduces the ndBMI (non-linear dynamic BMI), an algorithm designed to approximate any continuous force field, effectively mimicking the complex, non-linear control laws found in biological systems like spinal reflexes.
System Architecture and Methodology
The ndBMI operates through a closed-loop architecture that bridges the motor and sensory domains.
The system consists of two primary interfaces:
- Sensory Interface (Encoder): This component maps the state of the controlled device (e.g., its position $x$) into specific patterns of electrical stimulation delivered to the somatosensory cortex (S1).
- Motor Interface (Decoder): This component records neural responses from the motor cortex (M1) and decodes them into a "virtual point" ($x_v$) within the device's workspace.
The core innovation of this approach is the use of Multi-Dimensional Scaling (MDS). MDS is used to map high-dimensional neural spike trains onto a 2D physical workspace. By calculating the "distance" between different neural responses during a calibration phase, MDS creates a map where neural activity naturally corresponds to physical locations. The final force ($F$) applied to the device is then determined by the desired force field ($\phi$) at this virtual point: $F = \phi(x_v)$.
Key Results and Robustness
The researchers validated the ndBMI using synthetic neural data, testing its ability to approximate complex force fields and its resilience to noise.
The algorithm successfully approximated both Gaussian fields (useful for simple reaching tasks) and Dipole fields (necessary for complex behaviors like obstacle avoidance). A key finding was that performance improved significantly with the number of stimulation patterns; for instance, increasing from 8 to 128 stimuli drastically reduced position errors in the complex Dipole field.
Furthermore, the system demonstrated remarkable robustness. It maintained high accuracy even when 90% of neural selectivity was lost or when specific electrodes were simulated as malfunctioning. The study also showed that a user could exert "volitional steering" by modulating neural signals to shift the target equilibrium point in real-time.
Discussion and Limitations
By treating the BMI as an artificial reflex or musculoskeletal "platform," the ndBMI allows the brain to influence dynamical policies rather than just raw coordinates. This bio-mimetic approach reveals that the geometry of neural responses can be directly tethered to the geometry of physical tasks.
While promising, the study currently focuses on static, position-dependent force fields. Future work must address dynamic factors like velocity-dependent forces (friction/damping) and the co-adaptation process between the brain and the interface in live biological subjects. Validation in "wetware" (living animal models) remains a critical next step to confirm if real neurons can effectively navigate the MDS-based mapping.
Summary updated