The XBI is designed as a device for training and behavioral testing of rhesus macaques in their housing environment, and can also be used for environmental enrichment. It has been developed with five design requirements in mind. First, the device needs to be cage-mountable to allow easy access for the animals without human interference (Gazes et al., 2012; Richardson et al., 1990; Truppa et al., 2010; Weed et al., 1999) or having to restrain the animals during transportation to the setup. Second, the electronics and other internal parts need to be protected against dirt and spray water typically present in such environments. Third, the XBI must be robust to resist potential forces applied by the animals. Fourth, operating the device should be easy enough to be handled by different people, including non-scientific personnel. Finally, the XBI’s hard- and software should be flexible enough to allow for a wide variety of training procedures and experimental task designs. This includes complex visually instructed cognitive tasks with well-defined stimulus viewing conditions and a high degree of flexibility in how the animal interacts with the device.
To address these needs the XBI’s hardware is divided into two parts: the animal Interface (AI) and the control interface (CI) (Fig. 1). In the following, we will describe the main design features and technical specifications. More detailed information on custom-built parts or purchased equipment are available upon request from the corresponding author.
Animal interface (AI)
The AI, used inside the animal facility, is the part of the XBI to which the animal has access (Fig. 2). It consists of mechanical and electronic components. For handling and safety reasons, the mechanical parts are lightweight and, where possible, built from aluminum. The dimensions of the whole device are 106 cm × 93 cm × 30 cm (W × H × D) and it weighs approximately 23 kg. By reducing the size of the outer frame and using lighter panels, we expect to substantially reduce the weight of future versions. The AI can be stored or transported using a custom-built wheeled frame (Fig. 1A), providing comfortable access to the front and rear for cleaning and maintenance. The XBI can be used either with the cart (no lifting required) or by directly attaching it to the animal’s enclosure (freeing the cart). For safety reasons all electronics of the AI run on low-voltage (maximum 12 V). Parts close to the animal that have to be powered include the touch-screen as the interaction device, a peristaltic pump for delivering reward, a loudspeaker to provide feedback or instructions, a surveillance camera for remote observation, and a cable connector box to minimize the number of cables between both interfaces. The rest of the XBI electronics reside remotely in the CI.
All animals had access to the AI in their home enclosures. These consisted of a room-sized group compartment and a smaller front compartment, physically separable by a dividing gate. The AI is attached to the front compartment with an aluminum-mounting frame, replacing one side panel of the compartment (Fig. 2B). For nine out of 11 animals the front compartment was connected to the group compartment such that the tested animal could be seated on-sight with peer animals. For two out of 11 animals the arrangement of the front compartment with respect to the group compartment did not allow visual contact.
The middle part of the XBI-AI is shaped as a funnel that narrows to the dimensions of a touch-screen (ELO 1537L), such that only the 15-in. LCD display is accessible for the animal. The dimensions of the front opening of the funnel are 48.6 cm × 41 cm (W × H) and the distance to the screen is 26.2 cm. This distance was chosen based on prior experience with rhesus macaques interacting with a touch-screen in neurophysiology experiments in our laboratory (Gail et al., 2009; Westendorff et al., 2010). The display is operated at a resolution of 1024 × 768 at 75 Hz. The touch panel in front of the display utilizes ultrasonic waves in combination with piezoelectric transducers for the sensing of the touch signal with a positional accuracy of 2.5 mm or better. The touch-screen is designed to be resistant against mechanical forces. A stainless steel tube with 8-mm inner and 12-mm outer diameter reaches across the funnel, at a fixed distance of 24 cm from the touch-screen. Fluid reward is delivered through a 1-mm opening in a 30-mm spout in the middle of this tube, precisely controlled via a peristaltic pump (see below). The stainless steel tube with the spout can be rotated and adjusted horizontally and vertically in position. In this way it is possible to set it to comfortable positions for individual monkeys of different size. Given that the animals usually operate the device with the reward tube as close as possible to their mouths (Fig. S1), the eye-to-screen distance is around 28–32 cm, depending on an individual’s head orientation and size. The screen size of 30.4 cm horizontal and 22.8 cm vertical provides 54° of visual angle along the horizontal and 42° along the vertical axis.
The AI’s backside contains a reward unit consisting of a fluid container (2.5-L plastic bottle), connected to the metal reward tube using flexible PVC tubes with 6-mm inner diameter. These tubes are exchanged after every 2 weeks of use. A peristaltic pump (Verderflex OEM M025 DC) allows electronic control of the reward flow. This reward unit can be placed at either the left or right outer side of the funnel to adapt to different cage structures. The pump delivers 1.8 ml/s of activation time, with a precision of approximately 0.01 ml. The reward was precisely timed and dosed via the experimental control software, which is crucial for cognitive neuroscience testing.
A mono sound transducer (Visaton, SpeaKa 130 mm) is glued on the outside of one of the funnel walls, using the wall as resonator for sound amplification. A compact 160° wide-angle CCD camera (ABUS TV7512) with 480 TV lines (438 kPixel) resolution is attached to a small opening in the metal funnel, protected by a clear polycarbonate window. The wide-angle view enables monitoring of the monkey and of the video screen at the same time.
Except for the VGA video cable, all connections (including power and signal lines) are routed to the CI via a custom-made connector box and a standard parallel D-SUB 25 connector cable (up to 15 m). Thus, only these two cables have to be routed to the outside of the animal facility. Within the connector cable we used multiple leads for power and ground lines to increase the amount of current that can be delivered through the cable.
The overall maximal nominal power consumption for the AI is 37.6 W (touch-screen 22 W, camera 0.6 W, active peristaltic pump 15 W). With an operating DC voltage of 12 V the XBI draws a maximum nominal current of 3.13 A. In practice we measured a total current of 1.5 A.
The AI is build to be operated for years, even in a dirty and humid work environment such as an animal facility. The front side facing the monkey cage is resistant against feces, urine and direct water impact during cage cleaning procedures. On the backside of the AI all components are protected against spray water and particles larger than 2.5 mm. According to IEC 60529, the international protection marking level of the whole XBI is IP 33, with a substantially higher protection from the inside of the monkey cage.
The CI consists of all the hardware and software needed for controlling the AI. It usually operates from outside the animal facility, weighs 12.2 kg and fits into a transportable box (W: 59 cm, H: 12 cm, D: 38 cm) for easy transport. The CI receives and sends signals from the AI through the VGA and connector cables. A second custom-made connector box distributes all connections from the connector cable to the individual components. The VGA cable as well as the serial RS232 connection from the touch-screen is connected to a computer that controls the XBI (Fig. 1). To control various devices from the computer, we integrated a USB interface (Service USB plus, Böning und Kallenbach). This platform provides multiple analogue and digital GPIOs (General Purpose Inputs/Outputs) which can deliver currents of up to 1.3 A. One of the digital outputs is used for operating the peristaltic pump, while the others have not been used in the context of the experiments described here. In addition, the computer’s audio output is connected to a custom-built sound amplifier, which provides the audio signal for the sound transducer. The camera signal is routed to a video server (TRENDnet TV-VS1P) and from the video server to an analogue screen for on-site observation. The video server and the XBI computer are connected to the Local Area Network (LAN). In this way any computer on the LAN can be used for remotely controlling the XBI as well as recording videos and downloading data.
As long as the necessary interfaces are available, hardware requirements for the CI computer to run the XBI do not exceed those of standard desktop or laptop computers. We used VGA and USB connections with a RS232 adapter for the touch-screen in the AI, another USB port for the Service USB plus device, DVI-D for the CI’s screen, and the headphone audio out for the audio amplifier. Although LAN connectivity is not necessary for the XBI to operate, it provides useful remote control capability. The video server is not directly connected to the computer but can be accessed via LAN. For the computer we either used an Apple Mac mini (2.5 GHz Intel i5, 8 GB RAM) or an Apple MacBook (2.4 GHz Intel Core 2 Duo, 2 GB RAM). The Mac OS is used since it interfaces optimally with MWorks (http://mworks-project.org/). This open-source software is a highly flexible C++-based package for designing and real-time controlling behavioral tasks for neurophysiological and psychophysical experiments. MWorks can be expanded by dedicated software plug-ins to serve a wide range of experimental needs. Behavioral tasks are coded as XML files. A custom-made XML editor makes programming and modifying task files easy even for users without programming experience. MWorks runs in a client-server structure. The XBI can be run either as a standalone system or be operated via LAN. Data files are generated on the CI-computer that runs the server software.
Animals, grouping and fluid control
Overall, a total of 11 male rhesus monkeys (Macaca mulatta) were trained on the XBI within their housing facility. Three animals (Gro, Chi, and Zep) had access to the XBI as a group directly from the group compartment of their home cage. We report their behavioral data as group performance. We confirmed that an off-line analysis of the video footage allows for determining which animal was responsible for each of the XBI interactions. Since performance comparisons between individual animals are not the purpose of this report and since future ID tagging will render manual performance assignment to individuals unnecessary, we did not extend our pilot off-line analysis to the full data set.
The other eight animals had individual access to the XBI from within the smaller front compartment of their home enclosures. These eight animals were physically separated from their social group by a dividing wall separating the front compartment from the group compartment during the XBI sessions. Animals Fla, Alw, Nor, Odo, and Pru were in sight with their social group, while animals Han, Toa, and Zor were in sight only with members of other groups in the housing facility.
Most of the 11 animals had at least 2 h of unlimited access to water and fruits before and after each XBI session (Monday to Friday) and 24 h on all other days (see Table 1 for details). Two animals (Pru and Zor) were trained on the XBI under fluid control, in which the XBI provided the only access to fluid on working days (Monday to Friday). Animal Pru, in the early phases of the training, received plain water as reward. The other animals were rewarded with fruit-flavored sweetened water (active O2, Adelholzener) diluted with plain water at a ratio of 1:3.
Note that monkey Zor, a 12-year-old animal, was tested only during the development phase of the device.
To date four units of the XBI are in ongoing use and have been tested in various experiments. All experiments complied with institutional guidelines on Animal Care and Use of the German Primate Center and with European (Directive 2010/63/EU) and German national law and regulations, and were approved by regional authorities where necessary. Two experimental paradigms shall serve as examples of the functionality of the system and acceptance by the animals. The first paradigm, the accommodation study, probed the ability of naïve animals to autonomously learn how to successfully operate a touch-screen on a basic level with no formal training (e.g., training to human handling). The second experiment, the free-task selection tested the XBI as a cognitive testing system and as an enrichment tool.
Nine animals (age: 4–7 years) participated in the accommodation study (AS). They were naïve with respect to the XBI, and the accommodation study marked their first encounter with the device. Each animal had 90 min of daily access (typically from Monday to Friday) to the XBI over a period of 2 weeks excluding the weekend. None of the animals had previously participated in any type of cognitive training.
In the accommodation study the monkeys had to perform a simple touch task. At the beginning of each trial a steady blue (white for monkey Fla) square target stimulus 20 × 20 cm2, was displayed on the screen on a black background. Touching the target for at least 100 ms triggered a fluid reward (successful trial). Touching the background terminated the trial without a reward (unsuccessful trial). Each trial was followed by an inter-trial interval during which the screen remained black. After 1 s without touching the screen the next trial started. This requirement of releasing the touch of the screen prevented the animals from successfully completing a series of tasks by simply keeping a finger (or any other body parts) on the screen. In addition to the delivery of the fluid reward, two different sounds indicated whether a trial was a success or not.
One animal (Pru, 7 years old) participated in the Free-Task Selection (FTS). Note that before entering the free-task selection, the monkey underwent 4 months of positive reinforcement training to enter and exit the primate chair and 12 months of training on the XBI (see below for details).
In the free-task selection, at the beginning of each trial, four symbols were displayed on the screen (see Fig. 3), each one permanently associated with one subtask (Washburn et al., 1991):
The cross was associated with a simple touch-hold-release (THR) task, an extension of the touch task in the accommodation study. After the animal selected the cross symbol and after a 500-ms delay the four symbols were replaced by a gray square (5 × 5 cm). The animal had 4,000 ms to reach for the target, which once touched, it brightened. After 500–2,500 ms of maintaining the touch the square dimmed. Now the animal had to release the touch within 500 ms to successfully complete the trial. The position of the stimulus on the screen and the required hold-time were randomized trial-by-trial. For this subtask the average duration of a successful trial was 4.8 s from when the animal selected the cross symbol.
The rhombus was associated with a color-based delayed match-to-sample (MS) task. In MS trials the animal had to first touch a colored square (8 × 8 cm) at the center of the screen and after a randomized delay (1.5–3 s), touch the square with the same color amongst four differently colored squares of the same size displayed left, right, above, and below the screen center. The colors of the squares were randomly assigned trial-by-trial. The animal had to select the target within 4 s for correct performance, otherwise the trial would terminate without a reward. The same outcome would occur if the wrong stimulus was selected. For this subtask, the average length of a successful trial was 2.7 s.
The horizontal bar was associated with a random choice (RC) task in which the animal had to touch one of four identical 3 × 3 cm red squares that were randomly positioned on the screen. Only one randomly determined stimulus would trigger a reward. By setting the amount of reward to four times the reward in the touch-hold-release and match-to-sample tasks the average reward was equated across these task types. For this subtask the average length of a successful trial was 3.6 s.
The vertical bar was associated with a primate picture (PP) task in which one out of 20 photographs of non-human primates were shown on the screen for 5 s. After selection, no additional touch was necessary and no fluid reward was given in this task. For this subtask the average length of a trial was 5.6 s.
The animal was trained on the touch-hold-release task for over 6 months while technical aspects of the XBI prototype were under development and the match-to-sample task for 3 months. Once the monkey had reached a consistent performance above 80 % over 10 sessions (2 weeks) in these two tasks he was introduced to the free-task selection task. It included the two known tasks and the two novel tasks each associated with its corresponding symbol (see above). To determine the influence of relative reward amounts on relative choice probabilities, the first 31 sessions (3 months) of the free-task selection have been collected in two experimental conditions: lower reward RC task (20 sessions) versus higher reward RC task (11 sessions). We statistically verified the influence of relative reward amount on relative choice probabilities by the mean of the Multinomial Logit Model with estimated p-values using pairs cluster bootstrapped t-statistics (Cameron, Gelbach, & Miller, 2008).