Research over the last forty years has shown that the spinal cord, even when completely disconnected from the brain, retains the ability to produce rhythmic patterns of output that can control relatively simple actions such as walking. This capacity, first demonstrated in the cat, could not be demonstrated in humans. However, more recent evidence suggests that the human spinal cord can control leg muscles with walking-like patterns even when completely disconnected from the brain.¹ This discovery promises exciting advances in rehabilitative technologies for individuals with spinal cord injury.

Current research in therapy after spinal cord injury is studying the benefit of weight-bearing stepping on regaining motor function. In a typical setup, the injured patient is partially supported by a harness and held over a moving treadmill: therapists on either side of the patient assist with motion and foot placement. Although the patients can produce the basic pattern of stepping, they often lack the ability to pull up their foot prior to placing it at the beginning of a stride, a deficit referred to as foot drop. Assistance in this portion of their gait is provided by therapists, but it may be desirable to have an automated sensing-actuation system to provide this assistance.

(a) The force-sensitive resistor (FSR) used in this application. (b) Voltage-divider circuit with FSR, showing that power and data acquisition were provided by a laptop computer, and components were placed in a backpack allowing the user to move freely. (c) The central pattern generator (CPG) neuron received stride frequency as direct input, step time as a reset, and put out spikes in steady state with the desired frequency and phase relative to the step cycle (step times shown as vertical lines). (d) Future work would involve coupling the CPG neuron to an actuator capable of lifting the foot.

It is generally thought that neural oscillatory circuits in the spinal cord, called central pattern generators (CPG), underlie the production of rhythmic motor behavior such as locomotion.² To assist recovery, it may be desirable to build a hybrid circuit containing the injured CPG and artificial sensing, computation, and actuation elements to attempt to correct for CPG deficits after injury. During the 2004 workshop in Telluride, we considered the sensing and computation components of the artificial CPG (see Figure 1).

It has been shown that partial weight-bearing is extremely important for eliciting motor patterns in spinal-cord injured individuals, suggesting pressure input from the foot is a key component of normal locomotion. Taking our lead from this result, we used force-sensitive resistors (FSR, Interlink Electronics) placed on the bottom of each foot to measure contact pressure (Figure 1a). Output from the FSR was read into a computer for on-line processing (Figure 1b).

We used a single oscillating integrate-and-fire neuron model, with the stride frequency as its single input to the neuron, to simulate the CPG (Figure 1c). The goal was to use the CPG output, which was also at the stride frequency, to estimate the timing of ankle actuation so that foot drop could be avoided. It was, therefore, important to have the CPG oscillation at the appropriate phase of the step cycle. To accomplish this, the neuron had its voltage set to half the threshold value at the time of each foot strike. Since the neuron was tuned to have a voltage threshold of 1, this reset ensured that the neuron would fire a spike halfway through the step cycle. As with biological CPGs, the occurrence of such a spike could signal the actuation of a particular joint. Thus, since this circuit generates an artificial CPG that produces spikes at accurate points during the step cycle, its output could be used to actuate the ankle through external pneumatic actuators, and thus reduce the problem of foot drop (Figure 1d).

The artificial CPG neuron that we have studied represents the simplest possible implementation of a hybrid biological/neuromorphic CPG. It is likely that more realistic CPG model circuitry will allow the integration of more complex cues from the patient's gait pattern, and may be able to correct gait abnormalities more fully. It will be interesting to explore the potential of artificial CPG circuits, acting in parallel with their injured biological counterparts, for influencing plasticity and recovery in the injured nervous system.

2. M. Maegele, S. Muller, A. Wernig, V. R. Edgerton and S. J. Harkema, Recruitment of spinal motor pools during voluntary movements versus stepping after human spinal cord injury, J. Neurotrauma 19, pp. 1217-29 October, 2002.
   
   
3. V. Dietz, Spinal cord pattern generators for locomotion, Clinical Neurophysiology 114, pp. 1 August, 2003.