`Instruments and Gestural Sensors for
`Musical Interaction and Performance
`
`Joseph A. Paradiso
`
`MIT Media Laboratory E15-325
`Cambridge MA 02139 USA
`
`joep@media.mit.edu
`Tel: +1-617-253-8988
`Fax: +1-617-258-7168
`
`Keywords: Brain Opera, human-computer interface, multimodal input devices, electronic
`music interfaces, interactive music systems, percussion interfaces, interactive dance,
`electric field sensing, capacitive sensing
`
`ABSTRACT
`
`This paper describes the array of new musical instruments and interactive installations
`developed for the Brain Opera, a large, touring multimedia production, where the audience
`first explores a set of musical modes at a variety of novel, interactive stations before
`experiencing them in an actual performance. Most of the Brain Opera's installations were
`intended for the general public, employing different gestural measurements and mappings
`that allow an untrained audience to intuitively interact with music and graphics at various
`levels of complexity. Another set of instruments was designed for a trio of trained
`musicians, who used more deliberate technique to perform the composed music. This
`paper outlines the hardware and sensor systems behind these devices: the electric field
`sensors of the Gesture Wall and Sensor Chair, the smart piezoelectric touchpads of the
`Rhythm Tree, the instrumented springs in Harmonic Driving, the pressure-sensitive touch
`screens of the Melody Easels, and the multimodal Digital Baton, containing a tactile
`interface, inertial sensors, and precise optical tracker. Also discussed are a set of
`controllers developed for the Brain Opera, but not currently touring with the production,
`including the Magic Carpet (immersive body sensing with a smart floor and Doppler radar)
`and an 8-channel MIDI-controlled sonar rangefinder.
`
`____________________________________________________________________________________
`
`Version 2.0, November 1998; To appear in the Journal of New Music Research
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 1
`
`
`
`Figure 1: Overhead view of the Brain Opera Lobby truss structure, as being assembled
`before a Tokyo run in November 1996. All electronics are mounted atop the truss,
`leaving only the interactive interfaces (such as the Rhythm Tree bags at lower right)
`visible to the participants
`
`1) Introduction
`
`New sensing technologies and the steadily increasing power of embedded
`computation, PC's, and workstations have recently enabled sophisticated, large-scale
`experiments in interactive music to be conducted with the general public. Although most
`(e.g., Ulyate 1998) have been one-time occasions, the Brain Opera is the largest touring
`participatory electronic musical installation to have been thusfar constructed. The
`interactive section alone, termed the “Mind Forest” or "Lobby" (named after the Juilliard
`Theater's Marble Lobby where it opened in July, 1996 at the first Lincoln Center Festival),
`is composed of 29 separate installations, run by an array of circa 40 networked PC's and
`workstations. Figure 1 shows an overhead view of the Lobby electronics being deployed
`atop its supporting truss structure, indicating the large physical scale. During a Brain Opera
`run, these interactive stations are open to the general public, who wander through them
`freely, in any desired order. The stations are of 5 types, each creating a different
`experience and exploiting various gestural sensing and multimedia mapping modalities, as
`described in the following section. Some of these stations allow manipulation of sound
`structures, others acquire voice samples from the users, and others enable parametric
`manipulation of various Brain Opera themes. After about an hour of Lobby experience, the
`
`2
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 2
`
`
`
`audience is conducted into a theater space, where a trio of musicians performs the entire
`Brain Opera composition on a set of “hyperinstruments” (Machover 1991), similar in style
`and technology to those previously experienced in the Lobby.
`The Brain Opera, conceived and directed by Tod Machover, was designed and
`constructed by a highly interdisciplinary team at the MIT Media Laboratory during an
`intensive effort from the fall of 1995 through summer of 1996. A major artistic goal of this
`project was to integrate diverse, often unconnected control inputs and sound sources from
`the different Lobby participants into a coherent artistic experience that is "more than the
`sum of its parts", inspired by the way our minds congeal fragmented experiences into
`rational thought (Machover 1996). This congealing process was anticipated to culminate in
`the Brain Opera performance, where the diverse themes and activities experienced in the
`Lobby were actively sculpted into a succinct musical piece. Such analogies to brains and
`the thought process, particularly as interpreted by artificial intelligence pioneer Marvin
`Minsky (Minsky 1988), drove much of the initial Brain Opera inspiration, from the use of
`uncorrelated, essentially stochastic audience input (emulating neural stimulation) to the
`somewhat "biological" appearance of the physical set. More generally, the Brain Opera
`attempts to make a strong statement about actively involving non-specialized audiences in
`artistic environments, confronting many questions about interactive music to which ready
`answers are currently lacking (Machover 1996).
`The overall design of the Brain Opera as an interactive installation is described in
`(Orth 1997), and its artistic motivation and goals have been discussed in many articles;
`e.g., (Machover 1996), (Rothstein 1996), (Wilkinson 1997). This paper, in contrast,
`concentrates on the many different instruments and interactive stations developed for this
`project, describing their technical design, sensor architectures, and functional performance.
`The Brain Opera is by no means a fixed or purely experimental installation; it had to
`operate in many real-world environments (already having appeared at 7 international
`venues), and function with people of all sizes, ages, cultures, and experience levels. As a
`result, the interface technologies were chosen for their intuitiveness, overall robustness and
`lack of sensitivity to changing background conditions, noise, and clutter. This tended to
`rule out computation-intensive approaches, such as computer vision (e.g., Wren et. al.
`1997), which, although improving in performance, would be unable to function adequately
`in the very dense and dynamic Brain Opera environment.
`Most of the Brain Opera's software is run on IBM PCs under Windows 95 or NT
`using ROGUS (Denckla and Pelletier 1996), a C++ MIDI utility library developed for this
`project, although some of the performance instruments are based around Apple
`Macintoshes running vintage code written in Hyperlisp (Chung 1988). Most of the musical
`
`3
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 3
`
`
`
`Figure 2: Schematic of the Speaking and Singing Trees
`
`interfaces described in this paper were designed to communicate via MIDI. In order to limit
`data rates, continuous controllers were polled; i.e., an interface waits for a poll command
`(in this case, a MIDI program change directed to the appropriate channel), then responds
`with its latest data. All of the custom-designed interfaces employed a 68HC11-based
`circuit as their MIDI engine, incorporated as either a “Fish” for electric-field and capacitive
`sensing (Paradiso and Gershenfeld 1997) or “FishBrain” card. The latter is essentially a
`Fish without the capacitive sensing electronics; a 68HC11 with 4 raw analog inputs, 4
`adjustable (gain/bias) analog inputs, 4 digital inputs, 8 digital outputs, and MIDI plus
`RS-232 input/output. The FishBrain is used as a general-purpose MIDI interface to analog
`sensors. With minor modification, the same embedded 68HC11 code is run on nearly all
`the Brain Opera devices.
`
`2) The Lobby Installations
`
`The simplest and most plentiful Lobby stations were the Speaking Trees. As
`depicted in Fig. 2 and shown in Fig. 3, these interfaces feature a dedicated PC, pair of
`headphones, microphone, 10” color LCD screen, and a modified ProPoint mouse from
`Interlink Electronics (http://www.interlinkelec.com/). The ProPoint is a handheld device
`that allows the thumb to navigate the cursor by adjusting the center of pressure atop a
`fingertip-sized, force-sensitive resistor array; the “clicks” are still determined by a
`pushbutton (mounted for forefinger access). In order to accommodate the “organic” look
`of the Brain Opera, the ProPoint circuit cards were removed from their dull plastic
`housings and potted into a somewhat elastic, leaf-shaped mold. As seen in Fig. 3, these
`
`4
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 4
`
`
`
`
`Figure 3: Layout and photograph of a Speaking Tree
`
`components were all mounted onto an adjustable-height polypropylene frame that fit over
`the head, nicely encapsulating the user in a somewhat private and isolated environment. A
`simple 17" x 23" switchmat is mounted under the carpet beneath each speaking tree. When
`an occupant steps under the tree, the switchmat closes, bridging a set of polling lines on the
`PC serial port. When this event is detected, a MacroMind Director sequence starts up,
`featuring video clips of Marvin Minsky, whose Society of Mind (Minsky 1985) inspired
`the libretto and overall concept of the Brain Opera. Throughout the dialog, the image of
`Minsky asks the user several questions; their answers are recorded and indexed on the host
`PC, then subsequently transferred over the network to a bank of samplers in the theater for
`playback during following performances. There are a total of 15 Speaking Trees in the
`Brain Opera, 12 of which run different Director sequences. Although the dialog with
`Minsky can be interesting and amusing, it’s only one simple application of the facilities
`available at each Tree; several other, more engaging experiences are now being developed.
`More detail on the Speaking Trees can be found in (Orth 1997).
`Similar in construction are the Singing Trees, schematically shown at right in Fig.
`2. Lacking a tactile interface, they respond solely to the singing voice, which they analyze
`into 10 dynamic features. These parameters drive an algorithmic composition engine,
`which effectively resynthesizes the participant’s voice on a Kurzweil K2500 synthesizer.
`The Singing Trees look for consistency in the singing voice at a single pitch; the longer the
`pitch is held, the more tonal and “pleasing” the resynthesis becomes. The derived quality
`factors are also used to drive an animation playing on the LCD screen (Daniel 1996); as the
`pitch is held longer, the animation propagates forward and becomes more engaging (e.g., a
`ballerina appears and begins to dance, as in Fig. 2). When the voice falters, the animation
`
`5
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 5
`
`
`
`Figure 4: A Melody Easel (left) and Harmonic Driving joystick (right) in action
`
`
`
`rewinds into a set of simpler images. The audio and video feedback on the singing voice
`has proven quite effective; the tonal and visual rewards encourage even poor amateurs to
`try for a reasonable tone. There are 3 Singing Trees in the Brain Opera, each running
`different image sequences. More details on the Singing Tree design and synthesis/analysis
`algorithms are given in (Oliver, Yu, Metois 1997) and (Oliver 1997).
`Another relatively straightforward interface used in the Brain Opera is the Melody
`Easel. These are 21” computer monitors, recess-mounted into a hanging “table”, such their
`screens are horizontal and embedded into the tabletop (see Fig. 4, left). These monitors,
`however, are equipped with pressure-sensitive touchscreens (the IntelliTouch from ELO
`TouchSystems), which can deliver 11 bits of position accuracy and circa 5 bits of pressure
`information at 10 msec updates. Users manipulate a parametric sequence running one of
`the Brain Opera themes by moving a finger across the screen; the synthesized voices
`(generated on a Kurzweil K2500 sampler and Korg Prophecy synthesizer) respond to
`position, pressure, and velocity. A video sequence, playing on the monitor, is likewise
`perturbed by the finger position and pressure, using various realtime video processing
`algorithms (Dodge 1997). There are 3 melody easels in the Brain Opera. Each uses a pair
`of PC’s (one for music and another for video), and runs different musical voicings and
`visuals. Fig. 5 shows data from an IntelliTouch screen used in the Brain Opera for a finger
`tracing a circle and an “X”. The position and pressure data are shown vs. time at left, and
`as a raster (x vs. y) at right, with the pressure values determining the radius of the
`overplotted circles (higher pressure = wider circles); the pressure goes to zero when the
`finger is lifted off the glass. The IntelliTouch uses surface-acoustic waves propagating
`through the touchscreen glass to determine the finger coordinates; the timing of the acoustic
`absorption peak gives position and the amount of signal absorbed by the finger determines
`
`6
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 6
`
`
`
`Figure 5: Data from a pressure-sensing Melody Easel touchscreen
`
`Figure 6: The Harmonic Driving joystick and sensor data
`
`pressure (Kent 1997). Unfortunately, it does not work properly when more than one
`finger is in contact with the screen, but users quickly adapted to this drawback, and had no
`difficulty using this somewhat ubiquitous interface. As the analog signals produced by the
`screen always give response (the packaged digital controller produces data when only one
`finger is in contact), it is possible to modify the touchscreen so it detects and responds to
`multifinger input, albeit with some ambiguity, but still useful for simple musical
`interpretation.
`At the Harmonic Driving stations, the user "drives" an animated vehicle shaped like
`a note through a graphical and musical landscape. Rather than using a conventional
`steering wheel or commercial joystick, which would hint too heavily of a common arcade
`experience, the user controls the experience with a novel interface made from a large (2”
`diameter, 15” in length), bendable spring (Fig. 4 & 6), which has an entirely different feel,
`
`7
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 7
`
`
`
`Figure 7: Rhythm Tree string configuration (left) and pad schematic (right)
`
`
`
`better suited to the “jovial” mood of the music and nonphotorealistically-rendered graphics
`(Robertson 1997). Musical parameters are selected both graphically (by steering onto
`different “roads” or hitting musical objects) and continuously (the joystick parameters
`themselves are mapped directly onto musical effects). The two-axis bending angles are
`measured using capacitive sensing to detect the relative displacement between the spring's
`coils at its midpoint. As shown in Fig. 6, four pickups are mounted outside the coils at 90°
`intervals. These are made from small sections of the insulated copper conductors from
`heavy gauge electrical cable, tied and epoxied to the spring. A transmit electrode
`(broadcasting a 50 kHz sinewave), similarly fashioned, is wound completely around the
`coil above the pickups. As the spring bends, the pickups move closer together on one side
`(further apart on the other), and the capacitive coupling between transmitter and receivers
`changes accordingly. Shielded cables are run from these electrodes to a nearby
`preamplifier, then to a "Fish" electric field sensing circuit (Paradiso and Gershenfeld
`1997), which digitizes the four proximity signals into 7-bit MIDI values. The spring is
`electrically grounded to prevent extraneous coupling, and provided that hands are kept
`away from the pickups (as encouraged by the mounting geometry), the capacitive proximity
`measurement is quite accurate (Paradiso 1997a).
`The spring’s twist is also measured with a potentiometer that rotates through the
`relative angle between the top and bottom of the spring. The presence of a seated
`participant is detected when a light beam pointed across the chair is interrupted, at which
`point the experience is started; when the occupant leaves the seat, the software is
`automatically reset. The potentiometer and photodetector signals are also digitized by the
`Fish. Fig. 6 shows the resulting bend (difference between opposing pickups), twist, and
`occupancy signals for an occupant moving into the seat and putting the joystick through all
`
`8
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 8
`
`
`
`Figure 8: Photographs of a Rhythm Pad, showing top view (left) of unpotted circuitry
`with electronics, PVDF pickup and LED vs. bottom (right) with PIC microprocessor
`and bus connections
`
`
`
`degrees of freedom: first bending it around in a circle, then twisting it clockwise and
`counterclockwise, and finally bending it in horizontal then vertical axes before leaving.
`There are 3 Harmonic Driving units in the Brain Opera, all running the same experience;
`each uses a PC for music (generated by an E-Mu Morpheus synthesizer) and an IBM RS-
`6000 workstation for graphics. The hub of the joystick is embedded with an array of 8
`LEDs, which flash under MIDI control.
`The Rhythm Tree is the world’s largest electronic percussion rig, composed of 320
`smart drumpads, organized into 10 “strings” of 32 pads each. As schematically depicted in
`Fig. 7, the pads in each string are daisy-chained along an 19.2 Kbaud RS-485 serial bus,
`like a line of “Christmas bulbs”. Each pad contains a 16-MHz PIC 16C71 microprocessor
`from Microchip Systems (see http://www.microchip.com/), which monitors the bus,
`processes data from a piezoelectric (PVDF) sensor strip (Paradiso 1996), and controls the
`illumination of a large, on-pad LED. As seen in Fig. 8, all electronics, the PVDF pickup,
`and the LED are potted in a compliant, translucent urethane (Orth 1997), which is struck by
`the hand when the pad is played. The PIC digitizes the pad signal into 8 bits at 50 kHz.
`When a peak is detected over threshold, a valid strike is assumed, and the PIC extracts a set
`of features from a subsequent 0-15 msec remotely programmable interval of the pickup
`signal. These parameters include the polarity of the initial PVDF signal peak, the number
`of significant zero crossings detected, and the net integrated signal amplitude (producing 14
`bits of velocity information). The initial significant peak polarity yields a very reliable
`discrimination between top and side impacts, as illustrated for many separate hits
`superimposed at left in Fig. 9 (top hits start negative, whereas side hits start positive); this
`
`9
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 9
`
`
`
`Figure 9: Pickup response to different types of pad strikes
`
`arises from the PVDF strip initially going into compression or expansion. As shown at
`right in Fig. 9, the number of zero crossings detected over the sampling interval can
`discriminate between a sharp, elastic hit (where the pad exhibits strong mechanical
`ringdown), and a dull slap with the hand remaining on the pad (a heavily damped
`response). Because this parameter depends more on the pad construction and strike
`kinematics, it is less consistent for the player without introducing a longer sampling interval
`and excessive latency.
`Because they are all on a shared bus, every connected rhythm pad must have a
`unique ID. This can be accomplished in three ways for these pads. Each pad has a 100 W
`resistor in series between a line in the bus input and the corresponding line in the bus
`output; this signal also goes to one of the PIC A/D inputs. The concentrator sets this bus
`line to 5 Volts at the first pad, and the terminator holds this line to ground at the last pad,
`thus this sampled voltage, read after the PIC powers up and the lines stabilize, is
`proportional to the position of the pad along the chain (the resistors form a divider ladder).
`Although this has promise, when put into practice, problems crept in from voltage stability
`and component tolerance. Another A/D input pin of the PIC is fed by a solid-state white
`noise generator (Simonton 1973), enabling the pads to access truly random numbers, thus
`statistically accumulate different ID’s (Smith 1998). This works, but looses information on
`the physical position of each pad, which the music software desires. The technique that is
`currently used is very brute-force; each PIC on a string has a unique ID (running 0-31)
`programmed into its PROM.
`The drumpads employ a very efficient round-robin polling scheme that transfers
`data to the bus concentrator with minimal delay. After a drumpad has been struck and has
`hit data ready to send, it waits for a poll-setup message from the bus concentrator, with a
`
`10
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 10
`
`
`
`Figure 10: A typical audience encounter with a Rhythm Tree Bag
`
`
`
`base pad address as its argument. Starting at this base, the pads on a string are sequentially
`addressed after each subsequent serial bit sent from the concentrator. Each of these
`transmitted bits advances a counter in all pads; when this counter value matches the
`assigned ID of a pad with data waiting, that pad seizes the bus and responds with its data.
`The concentrator then transmits another poll-setup message commanding the pads to set
`their counters to the next pad in the sequence, and continues polling as before. As each pad
`is independently addressed, this scheme returns the hit data after a bounded, deterministic
`interval. Addressing sequential pads with a single transmitted bit (rather than a full
`address) entails minimal readout delay.
`The hit parameters are thereby passed to the concentrator, where they are formatted
`into a MIDI stream and routed through a MIDI merger (grouping up to 8 strings) to the host
`computer, which then triggers synthesized sounds and illuminates the pad LEDs according
`to a simple pre-defined mapping scheme with two sounds on each pad; one for top and
`another for side impacts (Back 1997). In order to facilitate easy testing using a commercial
`drum synthesizer, the pad number and high velocity byte are sent as a MIDI note, followed
`by the hit polarity, ringdown count, and low velocity byte sent as a pitch bend command.
`The drumpads generally respond within a 15 msec interval (much of which is due to the
`integration time rather than network latency); a bit slow for a performance instrument, but
`adequate for their application in the Brain Opera Lobby.
`The pad’s LED intensity is controlled by the PIC via duty-cycle modulation.
`Normally, the PIC is set in a mode that automatically illuminates the LED upon hit detection
`with an initial brilliance proportional to the detected velocity, then exponentially dimming.
`The LED can also be directly controlled or triggered over the bus (hence MIDI); this mode
`is exploited to send occasional “waves” of illumination across the strings.
`
`11
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 11
`
`
`
`Figure 11: Gesture Wall in action (left) and schematic (right)
`
`Nearly all pad parameters (trigger thresholds, integration times, LED modes, etc.)
`are programmable over the bus. A Visual Basic program has been written to allow these
`parameters to be rapidly adjusted for individual pads and groups of pads; a data file is then
`produced, which is downloaded to the pads by the music software. This parameter list is
`continually sent out over the bus in order to reprogram any pad processors that reset
`through power interruptions; a valuable feature, in that the RJ-11 connectors that mate to
`the pads are occasionally pulled and momentarily break contact (the initial, solid, linear
`structure of the rhythm tree design gave way to a less expensive mounting scheme using
`foam-filled bags, which have several mechanical disadvantages). Despite these drawbacks,
`the rhythm tree system successfully survives long periods of kinetic abuse (e.g., Fig. 10),
`as expected for such an installation in open, public venues.
`All pads also have a common bus line connected to the remaining PIC A/D input;
`each PIC can adjust the voltage on this line. Although it is currently unused, it was
`installed to enable fast collective communication and computation of a common function for
`other purposes, such as direct sound synthesis. Even though the PIC chips, having only
`64 bytes of data memory and 1K of PROM, provide a restrictive development
`environment, all PIC code was written entirely in C (http://www.ccsinfo.com/picc.html).
`The last installation in the Lobby is the Gesture Wall, which uses transmit-mode
`electric field sensors (Paradiso and Gershenfeld 1997) to measure the position and
`movement of a user's hands and body in front of a projection screen. The device is
`diagrammed at right in Fig. 11, and shown in operation at left; there are five Gesture Walls
`in the Brain Opera. A brass transmitter plate atop the floor is driven by a low-frequency
`sinusoidal signal (ranging 50-100 kHz; each gesture wall is tuned to a different frequency
`
`12
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 12
`
`
`
`
`
`Figure 12: Distribution of Gesture Wall transmitter drive voltage, corresponding to
`shoe impedance, for a sample of 700 Brain Opera attendees
`
`to avoid crosstalk) at very low voltage (2-20 Volts P-P), far removed from any FCC or
`health restrictions. When a performer steps atop the transmitter platform, this signal
`couples through the performer's shoes into their body. A set of four pickup antennas
`mounted on goosenecks around the perimeter of the screen are narrowly tuned through
`synchronous demodulation (Paradiso and Gershenfeld 1997) to receive this transmit signal
`and reject out-of-band background. The amplitude of these received signals, which
`corresponds to the strength of capacitive coupling (hence proximity) to the body, is
`detected and routed through log amplifiers to approximate a linear range-to-voltage
`relationship, then 8-bit digitized and output via MIDI to a PC running ROGUS. An LED
`potted in each sensor “bud” is driven with the detected signal, thus glows with increasing
`intensity as the performer's body approaches (see Fig. 11 left); these LED’s can also be
`directly controlled through MIDI to illuminate in any desired fashion.
`When transmitting into the body through the feet, the complex impedance of the
`shoe sole has a large effect on the coupling efficiency, hence signal strength. This depends
`not only on the size and thickness of the sole, but most heavily on the sole material and
`composition, thus varies highly from one shoe to the next, hence needs to be compensated.
`We solved this problem by having the player first put a hand flat on a calibration plate after
`stepping on the transmit electrode, measuring the current flowing into the body through the
`shoe sole. During the calibration interval, which requires well under a second, the
`transmitter's output voltage was servo'ed to drive this current to a fixed reference, thereby
`removing the effect of shoe impedance and making everybody more-or-less identically
`efficient transmitters. A small second-order nonlinearity was introduced into this loop to
`
`13
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 13
`
`
`
`Figure 13: Reconstructed Gesture Wall position for a hand tracing the indicated
`patterns
`
`compensate for the series capacitance of the insulated calibration plate. After calibration, all
`players can access the full range of response that the Gesture Wall is capable of delivering.
`Fig. 12 shows the distribution of transmitter drive voltages (corresponding to shoe
`impedances via the calibration servo loop) for all users of one Gesture Wall during a 1-
`week Brain Opera run in Lisbon, illustrating the associated spread in coupling efficiency
`caused by different shoes. As seen in the fitted curve, the data can be approximated by a
`gamma distribution (Hoel 1971).
`An attached PC determines the position of a hand in the plane of the receivers and
`its distance from this plane by linear combinations of the four sensor signals. The
`weighting factors are determined by a least-squares fit to data taken with a hand placed at 9
`predetermined locations in the sensor plane. This is done only once, after the Gesture Wall
`is first set up. The resulting data can track hand position very well, as seen in Fig. 13,
`which shows coordinates derived from the calibrated Gesture Wall sensors for a hand
`drawing in the air. As shown in the lower diagram, the hand first tracks a spiral, then a
`square, then both diagonals, with the “pen” changing every time the hand is pulled back.
`The individual shown in Fig. 11 is using the Gesture Wall in this “tracking” mode, with
`one hand forward and body back. More generally, people will introduce two hands at
`once, if not their entire body. In this case, the above algorithm produces “averaged”
`coordinates, which reflect the body’s bulk position and articulation. Although the least-
`squares calibration was performed with an average-sized person, the tracking accuracy will
`also vary somewhat with the size of the participant. The mappings chosen for the Gesture
`Wall were selected to nonetheless give adequate audio/visual response for a wide range of
`body types and postures. We have subsequently developed another device (Strickon and
`Paradiso 1998) based on a low-cost scanning laser rangefinder that can determine accurate
`
`14
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 14
`
`
`
`Figure 14: The Digital Baton and Gesture Tree in performance
`
`position of multiple hands in a plane, independent of body size or posture; although this
`was inspired by Gesture Wall applications, it is not currently in the Brain Opera.
`The “back end” of each Gesture Wall consisted of a pair of PC’s (one running the
`music and sensor analysis code and another running the graphics code), a Kurzweil 2500
`synthesizer, and a video projector. The musical mappings (Machover 1998) played
`sequences that would increase in amplitude as the body approached the sensor plane
`(starting at silence with the player away from the sensors), and change pitch range as the
`player moved their hands/body vertically (favoring low notes with hands near lower
`sensors, high notes with hands near upper sensors) while changing the instrument timbre
`and panning as the hands/body are moved from right to left. The visual mappings (Dodge
`1997), (Smith et. al. 1998) created perturbations to a video sequence (Daniel 1996) when a
`player approached the sensors, with effects centered at the perceived hand/body position.
`
`15
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 15
`
`
`
`Figure 15: The Brain Opera's Sensor Chair; diagram (left) and in performance (right)
`
`
`
`3) The Performance Instruments
`
`The segment of the Brain Opera that was actually performed before the audience
`was played by a trio, each using a different custom-designed electronic instrument. One of
`them, the "Gesture Tree" (visible at right in Fig. 14), is a simple hybrid of two of the
`interactive instruments described in the last section. A bag of Rhythm Tree pads enables
`percussive and tactile triggering of sounds, which are then modified in various ways by
`waving an arm around the four Gesture Wall sensors mounted above (the performer is
`standing on a transmitter plate). Although this instrument is usually played with bare
`hands, some performers have used metal drumsticks, to which the Gesture Wall pickups
`will also respond.
`Another of the instruments, the Sensor Chair (Fig. 15), is based solely on transmit-
`mode electric field sensing. It is very similar to the Gesture Wall, except here the
`performer sits on a chair with a transmit electrode affixed to the seat, providing excellent
`electrical coupling into the performer's body. Since the differences in shoe coupling (Fig.
`12) are no longer an issue here (everybody couples nearly as well through the seat; clothing
`differences have only minor effect), there is no need for a calibrator pickup. A linear
`software calibration, run roughly once a day for each chair performer, is sufficient to
`enable good, repeatable gesture response. A halogen bulb embedded in the hand receiver
`
`16
`
`IPR2020-00910
`Garmin, et al. EX1011 Page 16
`
`
`
`Figure 16: Schematic of the Digital Baton (right) and photograph (left)
`
`electrodes glows with proximity, providing visual feedback for performer and audience.
`As the performer is now seated, he is free to move his feet, which are likewise tracked with
`a pair of pickup electrodes mounted on the chair platform (lights underneath are similarly
`illuminated with foot position). A pair of footswitches are available for introducing hard,
`sensor-independent triggers for advancing mapping modes, etc. The chair electronics are
`thoroughly described in (Paradiso and Gershenfeld 1997). The chair system is extensive