The device (Figure 1) consists of three doors, three pellet dispensers, and three traps assembled onto a metal barrier, which is mounted on a linear slider, totaling 10 degrees of freedom. Each degree of freedom is driven by a pneumatic cylinder (20-40 psi) to eliminate electrical noise in neural recording. In an experiment, pellets are dispensed into the traps and then cleared away. If the metal barrier is at a forward position on the slider close to the monkey, and the door corresponding to that trap is open, then the monkey can obtain the reward.

A manifold of eight 4-way two-position solenoid air valves (Numatics 226-432B) is used to control the air cylinders. The trap cylinders are controlled by a single air valve, whereas every other cylinder has a valve dedicated to it alone. A 24V DC-power supply powers the manifold, and a single source of compressed air is connected to the input of manifold.

An array of 8 optoisolators (IXYS LCA710) mounted on a custom printed circuit board (PCB) is used to electronically control the valves. An optoisolator acts as a switch opening or closing the connection between the 24V power supply and a valve, depending on its input signal. Closing the connection activates the corresponding valve, changing the direction of air flow through it. The optoisolators take 5V signals as inputs, allowing them to easily interface to the GPIO pins of many micro-controllers (Arduino, Trinket 5V, Pinguino, Raspberry Pi, etc) as well as the National Instrument Data Acquisition (NIDAQ) system. Indicator LEDs at the input pins of the optoisolators light up when the corresponding valve is activated.

The pneumatic air cylinders are double-acting: they have air pressure ports on each end of the cylinder which force a plunger to either extend or retract by applying pressure to one port while venting the other in a controlled manner. The speed of application and venting of the pressure establishes the speed of the plunger movement. A rod is attached to the plunger that serves to transmit the motion to the mechanism it drives.

The linear slider consists of a rectangular aluminum plate (50cm × 11cm) mounted on a slider carriage riding on a modified box rail. The box rail itself is mounted onto the back side of a half-square tube (see Figure 2A). The double-acting cylinder is bolted onto the bottom side of the half-square tube with its rod attached to the back of the aluminum plate via an angle bracket.

A barrier made from five metal panels (10cm wide, 50cm high) is mounted on top of slider, which serves as a base to assemble the other components of the system. As the rod of the slider cylinder extends or retracts, the slider along with the metal barrier, slides along the rails. The cylinder plunger is extended by default and the barrier is approximately 35cm away from a monkey seated in front of the device. Air pressure and venting rate can be adjusted to select the slider speed and smoothness.

This entire setup is attached on top of a jack-stand that is bolted onto a square bracket base. The platform can be raised to a height of 50cm, allowing easy adjustments of the device for different experimental setups and monkey height. Each of the two side flaps of the bracket base is attached to the top of a magnetic base. Thus this apparatus is portable and can be mounted onto any metal surface.

On the bottom of the three center panel sections of the metal barrier are round holes (4cm diameter) through which a monkey can reach for pellet treats. The three center sections are arranged in a trapezoidal shape so they are equidistant from a monkey seated facing front-center of the barrier. A short round section of PVC (3cm length, 4cm diameter) shield is glued onto each opening. Air cylinders with the rods attached to aluminum plates (7cm × 9cm) are mounted vertically on the back of the barrier right above the holes. Activation of the cylinders would push the aluminum plate doors down to cover the corresponding holes.

Blocks of 9cm × 14cm × 3cm (width × height × thickness) ultra-high-molecular-density plastic (UHMD) are bolted onto the back of the three center sections covering the sliding doors (see Figure 2B). Slots are cut on the top side of the blocks against the barrier to let the door plate slide through. Holes of 4cm in diameter are also cut in the blocks corresponding to the locations of the barrier holes to be used as reward traps.

In our previous experiments, owl monkeys were well motivated by food rewards (Fitzsimmons, Drake et al. 2007; Fitzsimmons, An et al. 2009). For this reason, we incorporated a solid food dispenser in this system.

The pellet dispenser consists of Plexiglas tubes of 1cm in diameter mounted vertically on the back of the metal barrier, with pneumatically controlled gates that allow only one treat to be dispensed at a time. The pellets (0.26-inch Bio-Serv Dustless Precision Pellets for primates) are stacked on top of each other inside the tubes.

The dispenser tubes are inserted off-center into the top of the UHMD blocks. Horizontal channels of 1cm are cut across the blocks right above their hole cutouts. Pieces of 8cm × 1cm × 3cm (width × height × thickness) sliders are inserted into these channels, with holes of 0.5cm in diameter. Air cylinders are mounted on the UHMD blocks with the rods attached to one end of the sliders. A vertical channel is drilled from the center of the horizontal channel through the door cutout in the UHMD block to let the pellet drop down into the trap (see Figure 2C).

When the cylinder rod is fully retracted, the slider fits inside the horizontal channel completely, and the hole inside the slider is directly below the dispenser tube, accepting only the bottom-most pellet from the dispenser tube. When the cylinder rod is fully extended, the slider is pulled out such that its hole and the single pellet within are pulled to the middle of the horizontal channel. The pellet then drops through the vertical channel to a rest on top of the trap. The monkey can then reach through the open door-hole to access the treat.

Traps are used to control the pellet dispensing process. The rectangular UHMD block is cut horizontally through the axis of its door cutout and the bottom piece serves as a trap. Only one side of the trap remains attached to the back of the metal barrier via a hinge, allowing the other end to rotate away from the plane of the barrier. An air cylinder is attached to the top UHMD piece via another hinge perpendicular to the barrier, directly above the free end of the trap. The cylinder is coupled to a bolt on the free end of the trap such that the extension of the rod pushes the trap away from the barrier, while retraction brings it parallel to the barrier (see Figure 2D).

In the closed position, where the cylinder rod is retracted, the trap will catch any pellet deposited by the corresponding dispenser. When the trap is rotated away in the open position, any existing pellets will roll to the floor and it will not be able to catch any more pellets. All three trap cylinders are connected to the same air valve so they open and close collectively.

Infrared (IR) emitter-sensor pairs are mounted on the inside of the PVC shields in front of the doors to detect the monkey's reaching hand. The emitter is placed diametrically across from the sensor so the monkey's hand interrupts the IR beam. Both the emitter and the sensor are powered by a 5V power supply, and the sensor output can be sampled by NIDAQ or custom microcontroller digital inputs.

Light emitting diodes (LEDs) are attached to the back of the traps (Figure 1C) pointing upward to illuminate their corresponding door-holes. We limited the current through the LEDs to less than 20mA by connecting resistors to it in series so that the LEDs can be directly controlled by a NIDAQ without using the optoisolators to source the required current. If not using the NIDAQ, the current limit is constrained by the total current that the custom micro-controllers can source out of their I/O pins.

In our experiment, a customized PC (AMD Opteron processor 2.4GHz with 2GB of RAM) installed with Microsoft Windows XP was used to control the device. We have explored two ways for the PC to interface with the device.

First, the PC communicated to the device through a NIDAQ PCI card (PCI-6220) and connected to a connector block (NI SCB-68). The optoisolators’ input signals were active low and were controlled by the DAQ's digital output pins. The IR sensor outputs were connected to the DAQ's digital input pins. The LEDs were connected to the DAQ's digital output pins. Custom Matlab scripts were used to control the outputs to and record the inputs from the NIDAQ system. The advantages of this approach are the short installation time and short signal delay. The main disadvantages are the cost of the National Instrument hardware and its compatibility with non-Windows systems. The experiments described here used this approach for the short installation time.

The second method used an Arduino Uno micro-controller. The PC communicated with the Arduino through USB2.0. The Arduino's digital I/O pins served as the optoisolators’ and LEDs’ input control signals, and received the IR sensors’ output. Custom Arduino software were used to monitor the state of the pins and communicated with Python2.7 scripts on the PC for experimental control and data-recording.

Note that while many off-the-shelf micro-controllers are capable of such interface, we explored using the Arduino for ease of use, low price, and abundant community. The advantages of this approach are the order of magnitude hardware cost reduction, wide platform support, and high level of control protocol customization. The main disadvantages are the long setup time as custom software must be written for the Adruino and the PC, and the longer, more variable signal delay through USB2.0.

All surgical and behavioral procedures conformed to the National Research Council Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee.

Two owl monkeys, monkey A (female, 8 years old) and monkey P (female, 2 years old) were trained on a visually-cued arm reaching task. Experimental sessions were conducted during weekdays. To avoid hypoglycemia in the controlled food paradigm, animals were fed twice each weekday. They were given a small amount in the morning and then the majority of their ration in the afternoon. The morning ration was decreased to keep the monkey motivated during the experiments. After they returned to their home cages, the animals were given monkey chow and fruit. Daily records were kept of the amount of chow and fruit given. Any subject losing more than 15% of its ad libitum weight was temporarily removed from the study and given free access to chow, fruit, and jelly. In our experiments, both monkeys maintained stable body weight. We continually monitored the behavior and appearance of the animals for signs of stress or healthy issues. No such signs were observed.

On the weekends, the monkeys received fruit and chow ad libitum. On all days, the monkeys were given free access to water in their home cages.

Monkey P was not implanted. Monkey A had microwire electrode arrays chronically implanted (Kralik, Dimitrov et al. 2001): 4 × 8 arrays in the left primary motor (M1) and left posterior parietal (PP) cortices, and 2 × 8 arrays in both the left and right primary somatosensory (S1) cortices. The chronic electrode assemblies were prepared in-house from 50 μm Teflon-coated steel microwires, paced 300 μm apart.

Before the naïve monkeys were introduced to the task of interest, they were first trained to reach through the open doors for pellets dropped at random, with no regard to number of attempts. Once they learned where to obtain treats, the slider m between reaches, with speed adjusted to not scare the monkeys. After the monkeys became accustomed to this set up, light stimulus was added. The LED corresponding to the pellet location was turned on indefinitely until the monkeys reached for the correct door and obtained the treat. After about 5 to 10 such experimental sessions, the monkeys were then started on the visually-cued experimental task training. During these preliminary sessions, the monkeys were constrained to the chair for a progressively longer time, up to 2 hours.

The task of interest for our experiments was a visually-cued task (Figure 3). The experimental setup was housed in a sound-proof room, with very dim red light since the Aotus is nocturnal and falls asleep in bright lighting conditions. The monkeys were seated in a primate chair (Crist Instrument Co. polycarbonate primate chair for New World monkeys) facing front-center of the apparatus. The monkeys were restrained inside the chair with a soft, figure-8 harness.

Each trial of the task started with the barrier away from the subjects. The control program randomly chose one of the three reward locations, and the corresponding LED lit up for 2 seconds. The door stayed illuminated for another 2 seconds as the slider brought the metal barrier forward into the monkey's reach, at which time the reach sensors became active. When the IR sensors detected a correct hand reach, a pellet was dispensed for the monkey into the corresponding trap. Five seconds after any reach detection, traps were activated to clear any pellet residues and the barrier slid back to the start position. A new trial then started after a 2 second inter-trial interval.

To prevent the monkey from developing a spatial preference towards a particular door, correction trials were imposed as trials following incorrect trials, with the target location unchanged. A treat was given for correct performance on correction trials, however, and an incorrect reach in a correction trial would results in no treat and another subsequent correction trial with the same target location.

Monkey A spent 20 sessions over 35 days getting used to the device following the procedures described above before starting the experimental task. Monkey P spent 23 sessions over 40 days before starting the experimental task.

Each visually-cued task session lasted between 60-90 minutes, depending on animal motivation. The total number of trials for each session was between 50-100, including error trials and correction trials. Monkey A performed five sessions of 100 trials each for this task, of which 78.6±7.6% (mean±standard deviation) were random target trials, and 21.4±7.6% were correction trials. The session performance, measured as the percentage of correct reaches toward random targets, increased over the sessions (Spearman rank correlation test, p=0.030), eventually reaching 92.3% accuracy (Figure 4). Monkey P performed 32 sessions of 81.4±19.7 trials (mean±standard deviation) each for the task, of which 66.0±14.2% were random target trials, and 34.0±14.2% were correction trials. Its session performance increased over the sessions as well (Spearman rank correlation test, p=4.09e-5), reaching a maximum of 90.1% accuracy (Figure 4). These steady improvements to consistently high performance demonstrate that the monkeys have learned to recognize the association between the light stimulus and reward location.

In 85% of all trials, the time between the barrier stopping and the monkey's hand reaching the target was less than 2 seconds. The recorded videos showed that the preemptive reaching attempts accounted for most incorrect trials in the later sessions. The monkeys would sometimes reach preemptively for the door before the slider stopped and reach sensors activated. Upon receiving no reward, the monkey would reach for a different door. At this point the reach sensors were activated and consequently detected a wrong reach, resulting in an incorrect trial.

The device functioned reliably, with malfunctions associated with the dispenser mechanism occurring roughly once every 20 sessions. These were usually due to pellet breakage obstructing the horizontal slider or the vertical channel. The blockages were cleared easily by poking the slider channels with a thin wire.

Following initial behavioral training (Fig. 4), we started systematic neural recording in monkey A. Here we present neuronal data from four recording sessions. In one session, the behavioral task was the same as during training. In the following three sessions, the task modification was introduced: the light cue was turned on only after the slider stopped in front of the monkey. All the details of these neurophysiological data will be addressed in our future publications. Here we only report directional tuning properties of M1 neurons for the reach and grasp period of the task.

The color plot of Figure 5a shows perievent time histograms (PETHs) for 35 M1 neurons recorded in a representative session. In this plot, horizontal lines represent the trial average PETHs for each neuron. The PETHs were calculated by first dividing the entire session into 10ms time-bins, which were used to determine the time-bin mean and standard deviations. Each time-bin was then normalized with those statistics to obtain normalized firing rates. The PETHs were centered on the time when the monkey's hand reached the target. Clearly, the neurons exhibited different patterns of activity when the monkey reached for different doors, i.e. they were directionally tuned (Georgopoulos, Schwartz et al. 1986).

To quantify directional tuning of each neuron, we calculated average firing rate for the interval −300 to 100ms relative to the time of reaching the target. Average rates were calculated individually for each trial and then entered into Kruskal-Wallis analysis to assess statistical significance of directional tuning. On average, 47.9±6.3% (mean±standard error) of the M1 neurons were significantly tuned in the sessions analyzed.

Preferred direction was assessed as the target for which the firing rate was maximal for the analysis interval. Figure 5B shows neuronal preferred directions for all recorded neurons for the same session as in Fig. 5A. Figure 6 shows the average fraction of neurons by preferred direction. 42.1±5.5% of the M1 neurons preferred the contralateral target, 26.4±7.6% preferred the center target, and 31.4±2.6% preferred the ipsilateral target.

Thus, our experimental apparatus permitted automation of a reach-to-grasp task, and could be combined with neurophysiological recordings to assess the directional tuning properties in a population of cortical neurons.