|Welcome to The Neuromorphic Engineer|
|Pinning the tail on the human|
PDF version | Permalink
Ever thought that a prehensile tail would be useful? The potential applications span from enhanced physical agility to social expression to a simple 'helping hand'. A longtime dream of mine has been to construct a robotic tail controlled by signals from electromyography (EMG) using skin-surface electrodes. EMG-based control of prosthetic limbs has a long history and is being developed for future tele-operated robotic actuators for use in space and other hazardous environments. Because EMG measures muscle activation, the signal represents the force that the muscle is exerting on the body and the environment.
At this summer's Telluride Neuromorphic Engineering Workshop, a team of enthusiastic participants decided to attempt a proof-of-concept project, demonstrating a simple two-channel (four-state) tail control. Our project consisted of three components: EMG signal detection, signal processing/command recognition, and the motorized tail. Our team consisted of: Pamela Abshire (University Maryland), Chris Assad (Jet Propulsion Laboratory), Rodrigo Alvarez (University of Pennsylvania), Lena Ting (Georgia Institute of Technology), Nima Mesgarani (University of Maryland), and Massimiliano Giulioni (Italian Institute of Health).
EMG signals measured on the skin surface over an active muscle can be as large as a few millivolts in amplitude, with frequencies mainly between 20-400Hz. They are typically measured with two skin electrodes placed along the length of the muscle and a high-input-impedance differential amplifier. A low-impedance ground electrode is placed on the body surface away from the muscle to control the common-mode voltage for the amplifier. Most clinical (commercial) systems use standardized wet electrodes (Ag/AgCl), each with an adhesive patch to hold it and the conductive gel securely against the skin surface. While we successfully designed and tested our own amplifiers, we ultimately used two commercial amplifiers that had better noise characteristics and could be connected directly onto the electrode patches. Figure 1 shows example signal data from a forearm muscle being rapidly twitched. Again of about 1000 was used here. In our final version, two sets of electrodes were placed vertically over the lower back muscles (the erector spinae muscle group) about 1.5in (4cm) from the spine.
Following analog amplification, we sampled the waveform using a multi-channel Measurement Computing USB (universal serial bus) analog-to-digital converter (ADC) at 500 samples/sec in 100ms blocks. This USB device included digital outputs as well. A laptop computer running MATLAB was used to control data acquisition and provide software control. The waveforms were subsequently rectified, low-pass filtered, and compared against a level threshold. We now had two digital channels that indicated when the left, right, or both back muscles were contracting.
To test the feasibility of controlling a robotic device using the EMG signals, we built a prototype robotic tail. Its core consisted of a light, flexible, steel cable that was easily bent yet resisted compression. This cable was fitted with eight circular flanges through which three parallel strings (Spectra Cable 0.030in) were threaded through holes on the edges. Applying tension to one string effectively bends the assembly as the effective length of this string reduces while the length of the steel cable remains constant forming an arc. The three tensional strings were separated by 120° along the circumference of the guide to control the tail's two degrees of freedom. Tension was applied to each string using the shaft of a Solarbotics motor as a winch mechanism. The motors and the steel cable were both mounted to a base plate that served as the base of the tail. The default gear ratio was reduced by removing a gear stage to facilitate passive back-driving of the motor.
The three motors were driven by one of three possible EMG commands: a left muscle contraction drove the left motor to pull, a right muscle contraction drove the right motor to pull, and co-contraction of the back muscles drove the third (downward) motor to pull. The three motors were connected in a star configuration with a common node in the center. When one motor was driven in the pulling direction, the current flowed through the motor towards the center of the star and then outward through the two other motors in the releasing direction. The pulling motor bent the tail and the other motors weakly released their strings. This system, while clever, suffered from both weak motor strength and frictional imbalances producing either too much or too little tension.
The final system (see Figure 3) was ‘portable”, consisting of a laptop, USB-based ADC, batteries, and a tangle of wires. The tail mechanism was mounted on a commercial lower-back support product. The tail would wag naturally while walking, due to the alternating muscle activations with each step. Leaning forward or deliberate stomach-muscle contractions produced back muscle co-contractions that pulled the tail down. Though it worked well, the lack of proprioceptive or visual feedback made tail control somewhat confusing.
This project initiated new ideas about how to use many more electrodes and signal-separation techniques to resolve the signals of muscle subsets that are currently blurred together on a single set of electrodes. The signal-to-noise ratio of the back-muscle measurements were quite high, which allowed a close look at the complex muscle activations that underlie balance and posture in humans.
Tell us what to cover!
If you'd like to write an article or know of someone else who is doing relevant and interesting stuff, let us know. E-mail the editor and suggest the subject for the article and, if you're suggesting someone else's work, tell us their name, affiliation, and e-mail.