Play That Trunky Music: Development of an Auditory Enrichment Device for Elephants in Zoos
Authors: 
Arianna Mastali, Benjamin Mayo, Charles Ramey, Nate Elgart, Kirby Miller, Melody JacksonAuthors Info & Claims
Arianna Mastali, Benjamin Mayo, Charles Ramey, Nate Elgart, Kirby Miller, Melody JacksonAuthors Info & ClaimsArticle No.: 7, Pages 1 - 14
Abstract
Elephants are a quintessential animal for zoos and wildlife parks. These zoo-housed animals serve as ambassadors to educate the public about their respective species as well as wildlife more broadly. As with any zoo-housed animal, physical and cognitive engagement and exercise are crucially important to the well-being of zoo-housed elephants. A key component of cognitive stimulation for elephants is a complex and variable environment. We designed and deployed an instrumented enrichment device for African elephants (Loxodonta africana) at Zoo Atlanta, augmenting their existing food-based environmental enrichment with audio cognitive enrichment. To gauge elephant interest in our device, we compared usage of the existing food-based enrichment before and after augmentation with audio. The device was installed for 7 days and 10 hours and had a positive impact on frequency and retention time with the existing enrichment, increasing frequency of usage by 81 instances and retention time by 3 hours, 28 minutes, and 23 seconds. While our audio enrichment device was successful at collecting data with 88.14% accuracy, improvements could be made to the sensing methods to reduce the rate of false actuations. Overall, the study is an example of successfully collecting longitudinal data with elephants and showed that these elephants responded positively to sound enrichment.
Figure 1:
1 Introduction
For many years, zoos have researched the key factors affecting elephant welfare in efforts to provide the best care and help maintain a healthy physical and mental state [12, 23, 32]. After studying the effect of factors such as housing or social environment on welfare, for example, Greco et al. found that stereotypic movements such as pacing (walking back and forth) could be lowered by “enhancing their elephants’ social environment and the spatial complexity” [12]. The importance of complexity was also reinforced by a study that showed a complex environment had a higher impact than habitat size on activity levels for zoo elephants [28]. As those responsible for planning zoo-housed elephants’ daily lives, keepers are tasked with providing stimulating enrichment devices, toys, browse (pieces of bamboo, trees, or shrubs), and exercise routines in order to meet the demanding physical and mental stimulation needs of these intelligent creatures.
In this project, we designed, built, and implemented an interactive enrichment device for the African elephants (Loxodonta africana) at Zoo Atlanta that allows them to play low-frequency sounds as they insert their trunks into an enrichment wall. We chose low-frequency tones as elephants often communicate through low-frequency sounds in the wild [29]. By introducing interactive elements into their habitat, our device provided these captive elephants with a novel activity that promoted both cognitive and physical activity. We show that the device had a significant impact on the number of instances the elephant approached the enrichment wall. The device also had a positive impact on the duration of time spent interacting with the wall, albeit not a significant one. Additionally, we show that the sensing system was able to detect a trunk’s presence, but its performance could be improved by exploring the use of alternative sensing mechanisms, as obstacles such as hay caused several instances of false accusations.
2 Background
In order to ensure the utmost care for elephants, resources are in place to guide keepers and organizations alike. The Association of Zoos and Aquariums (AZA) provides care manuals, used by accredited zoos, which include enrichment guidelines that aim to stimulate and maintain both the physical and mental health of animals [3]. Several evaluation tools exist to measure behavioral welfare, such as the one designed and tested by Yon et al. for example [34]. These evaluation tools measure welfare through numerous physical and behavioral measurements, as laid out by Williams et al. [32]. Furthermore, organizations exist for keepers to collaborate and share resources to improve well-being, such as the Elephant Managers Association (EMA) [2]. This organization publishes seasonal journals with content supplied by members documenting elephant care developments across a number of zoos for community awareness. Additionally, the organization provides a database of enrichment resources (including vendors, toy ideas, material sourcing, research articles, and regional browse) and community-gathered care products with vendor sources. While it is infeasible for habitats at zoos and wildlife parks to furnish tens of square kilometers to provide spatial resources similar to the wild, these organizations can still provide activities that encourage physical and mental activities for elephants. Additionally, providing access to enrichment devices and activities can give more independence to those animals through choice and control, leading to better welfare [6].
3 Related Work
In this work, we leverage findings from prior research focused on animal-computer interaction and interactive zoo enrichment systems to build a unique enrichment system for zoo-housed elephants at Zoo Atlanta.
3.1 Animal-Computer Interaction
Animal-computer interaction (ACI) research gained sufficient momentum to become a field in the early 2010s as researchers from a variety of academic fields began to publish works that concentrated on developing computing systems for non-human users [18]. Through wearable and interactive technologies, humans can better understand an animal’s behavior and from that, provide better care for the animal. Recent ACI studies have focused on designing wearable systems for gait detection in cattle as well as canines to detect lameness and reduce orthopedic injuries [13, 26]. A paper by Byrne et al. explores ideas for systems to monitor stress and well-being of bears in zoo habitats. The proposed solutions ranged from video observations to wearable monitoring devices [4]. In addition to on-body sensors, Byrne et al. introduced to the ACI community the idea that behavior could be quantified from interaction with instrumented enrichment devices by creating dog toys fitted with sensors and other electronics [5]. While the body of prior work in animal welfare monitoring is growing, Diana et al. state in their survey paper that little of that work pertains to zoo animals [8]. They also point out that even among the existing studies, most do not involve automated detection algorithms. We designed our elephant enrichment device to utilize sensor data to autonomously monitor behavior.
3.2 Interactive Zoo Enrichment
Enrichment is a crucial part of a zoo-housed animal’s well-being, and within the field of animal-computer interaction, researchers have been exploring new and creative ways to incorporate technology into enrichment practices. Before the official establishment of ACI as a field, Markowitz was revolutionizing enrichment with environmental devices, several of them incorporating technological systems [19]. One of these systems involved an acoustic prey device for leopards that rewarded the leopards with food when they approached the system after bird noises were emitted from speakers [20]. In general, many of these existing studies have focused on the design and effect of touchscreen systems for primates. For example, Martin et al. designed a touchscreen system that assesses chimpanzees’ abilities to perform matching tasks [21, 22]. Others have conducted studies co-designing with orangutans for a digital enrichment installation and measuring the effects the system has on the orangutans who interact with it, as well as the visitors who observe the interactions [25, 31]. With regard to non-touchscreen enrichment devices, studies have prototyped physical button interfaces for saki monkeys which resulted in guidelines for physical prototyping for non-human users. Those same researchers designed an interactive system for saki monkeys which gave the monkeys a choice between audio and visual stimuli. The authors observed how the opportunity of choice affected the animals’ usage of their interactive system [14, 16]. Another study involved the design of an interactive audio device that lemurs could activate using infrared sensors. They showed that while the lemurs all preferred audio over silence, they each had individual preferences based on which noises they activated and how long they interacted with the system [15]. Branching away from primates, researchers worked with the San Diego Zoo to design two sound-based enrichment devices for a hyacinth macaw, allowing the bird to have the control to listen to music when desired [17]. Like these prior studies, our goal was to provide the elephants in our study with more control over their environment through the design of a technological system in order to observe their responses to sound enrichment.
3.3 Human-Computer Interaction for Elephants
Previous research for elephant-specific computer interaction has primarily explored novel enrichment methods enabled by interactive electrical systems. The works of French et al. established a foundation of research investigating material usage, interaction methods, sensor types, and depth of elephant cognitive association to ultimately aid in creating interactive devices that provide the elephants with more choice and complexity in their environment [10, 11]. This work also provides a basis for understanding how a user-centered design process can be applied to an elephant. The devices prototyped in these studies provide enrichment through tactile and acoustic feedback and were reported to provoke interest and positive feedback from the elephants during testing. Results from the acoustic enrichment device reported that frequencies around 60 Hz provoked the most interest from testing [11]. These enrichment devices, however, were routinely broken by the elephants and were unable to be utilized for longitudinal data collection. Another study developed an interactive auditory enrichment device that encouraged play behaviors by playing a variety of sounds depending on the orientation of the device as it was flipped [33]. Results showed that elephants demonstrated an understanding of operating the device, and overall positive reactions to the device, measured through increased interaction when compared to passive objects, as well as increased exploration behaviors.
While previous research has supported using technology to facilitate novel enrichment methods, our aim was to evolve the design of interactive devices and employ embedded electronics to provide longitudinal engagement data collection for the participating elephants. Due to elephants’ strong sense of hearing and vocal tendencies, we designed an interactive acoustic enrichment device augmenting their existing food-based enrichment with low-frequency sounds. As reported in these previous studies, elephants can easily destroy almost any presented device, which in a sense provides enrichment of its own. However, this encouraged us to set a goal of designing a system that could withstand elephant interaction, and provide longitudinal data for a comprehensive analysis representing overall engagement over time.
4 Methods
4.1 Ethics Statement
This study was approved by Georgia Tech’s IACUC protocol JACKSON-A100685U as well as Zoo Atlanta’s research committee. The research procedures with the human participants in this study were approved by Georgia Tech’s IRB protocol H23347. Every iteration of our device was first approved by the head keeper in order to maintain a high standard of safety for the elephants throughout our study. Additionally, as we wanted to give the elephants more control and agency, we let the elephants approach the wall of their own volition.
4.2 Participants
We designed our audio enrichment device for three African elephants at Zoo Atlanta. The three elephants consisted of one male and two female elephants, ages 36-41. The species, age, and sex of each elephant can be seen in Table 1.
Table 1:
| Participant | Species | Age | Sex |
|---|---|---|---|
| 1 | African elephant (Loxodonta africana) | 36 | Male |
| 2 | African elephant (Loxodonta africana) | 41 | Female |
| 3 | African elephant (Loxodonta africana) | 41 | Female |
4.3 Research and Ideation
4.3.1 Qualitative Research.
During our initial meeting with the elephant keeper team, we engaged in a contextual inquiry followed by semi-structured interviews, learning more about the keepers’ daily tasks with the elephants. Our primary goal was to gather insights and better understand the considerations the keeper team takes for each specific elephant in their care. Specific considerations, such as the existing enrichment routines, were utilized to develop our project direction. The contextual inquiry consisted of our team shadowing the keepers as they conducted a morning exercise routine with each member of the elephant herd. During the exercises, the elephants demonstrated trained husbandry behaviors used in standard care routines to evaluate their health [Fig. 2]. The following semi-structured interview involved a series of questions that were curated to better understand the establishment of the herd, the elephants’ individual personalities, enrichment methods, safety considerations, and factors for building around the existing habitat.
On a later day following the initial session, we conducted an observation for approximately 3 hours. During the observation, we took notes on each elephant’s activity, their various methods of getting food or enrichment, and their interactions with their habitat and with one another.
Figure 2:
4.3.2 Concept Development.
Informed by our review of prior literature and qualitative research, we engaged in an ideation session to explore potential enrichment device designs. The design ideas were guided primarily by the affordances they would offer the elephants for interaction, as well as the functional and non-functional requirements listed in Table 2 that came out of our qualitative research. Second, given potential designs, we narrowed down the ideas based on what we believed could be fabricated given space and budgetary constraints. We also considered which of the potential designs would be best suited for the existing habitat setup and individual elephant preferences. From our semi-structured interviews with the elephant keeper team, we knew these particular elephants have a high food drive and prefer enrichment activities that incorporate food-based reinforcement. However, we were curious to explore the elephants’ interest in other methods of sensory enrichment to add complexity to the existing habitat. After our ideation session, we settled on three ideas to develop sketches for, and present to the zoo for a final selection:
(1)
An elephant-controlled heater placed in the elephant barn, a separate indoor space that the elephants are put in during cleaning times and when the weather is too cold or rough. This would allow each elephant to control a personal heater using a push button via trunk interaction for comfort during cold winter days.
(2)
An interactive wind chime that would use capacitive touch for elephants to play various tones by touching each element with their trunk.
(3)
A wall perforated with holes for an elephant to stick their trunk through, with different holes providing different forms of enrichment between possible acoustic, olfactory, and food enrichment.
Table 2:
| Functional | Nonfunctional |
|---|---|
| Durable, unable to be destroyed by an elephant | Assists keepers in an aspect of normal routine |
| Electronics are fully contained | Gives choice/agency to elephants over habitat |
| Design blends with nature | Provides an opportunity for interaction with their trunk |
| Placement of the device must be visible | Provides mental stimulation |
| Placement of the device must be accessible to their trunk | Logs engagement |
| Usable at all hours |
After meeting with the zoo to discuss the three potential ideas, we decided to move forward with the enrichment wall device. We favored the idea of the enrichment wall because it follows an existing enrichment model (the current enrichment wall installed in the habitat), would require the least amount of additional infrastructure, and minimizes elephant exposure to electrical elements. A sketch of the design idea is shown in Figure 3. In addition to the food reinforcement given by the existing barrels of hay behind the wall, we selected acoustic reinforcement instead of olfactory, as a speaker can be placed further away from the wall, avoiding direct interaction between the elephants and electronics.
Figure 3:
Figure 4:
4.4 Fabrication and Installation
4.4.1 Physical Prototyping.
To equip the enrichment wall with an interactive interface that plays sounds as elephants reach their trunks through, we decided to install sensors inside the holes in the wall. To facilitate the installation of sensors on the wall, without drilling into the wall to install fasteners, we decided to put short lengths of polyvinyl chloride (PVC) in the holes to position and protect the sensors. To ensure the proper fit of our device into the wall, we used the photogrammetry software Reality Capture to generate a geometric accurate 3D model of a portion of the enrichment wall derived from a library of photographs taken on-site [27]. This allowed us to match the curved geometry of the wall, as shown in Figure 4. We then used this model to create a computer numerical control (CNC) milled form out of medium-density fiberboard (MDF) as the negative geometry of the hole bevel to which we vacuum-formed high-impact polystyrene (HIPS) sheets, creating a beveled ring piece that sits flush against the concave bevel surrounding each of the holes [Fig. 5]. A section of 8” diameter PVC pipe cut in length to the thickness of the wall formed the structure of the insert, the HIPS bevel ring was placed around the PVC, then capped off with a laser-cut acrylic faceplate to create a plug on one end, ensuring the device cannot be pulled through the hole to the habitat side of the wall [Fig. 6]. This plug end had a cavity created by the connection between the HIPS bevel ring, PVC, and acrylic faceplate that housed the electronics and was securely attached with construction adhesive to ensure sufficient waterproofing on the exterior face and bar elephant access. We additionally CNC routed a pink foam model of the wall to test the prototype fit without the need to travel to the zoo each time.
Figure 5:
Following in-house prototype testing, we abandoned the bevel inserts in favor of a less intrusive solution with time-of-flight (ToF) sensors placed on the back of the wall away from potential direct trunk contact and pointed directly down, avoiding issues with dirt blocking the sensors due to extended usage. The new device consisted of T-junction PVC waterproof conduit boxes, cross-linked polyethylene (PEX) tubing, and ratchet straps while using an identical electronics setup as the previous system. A ToF sensor and microcontroller were placed within each conduit box, with the longest edge of the box placed parallel to the ground. This allowed the power and communications wires to pass through each box and into the conduit sections which linked the boxes together. The short perpendicular opening of the T-junction pointed across the opening of the hole, which allowed the ToF sensor to detect when trunks were inserted through the hole [Fig. 7].
Figure 6:
Figure 7:
4.4.2 Electronics.
The electronics system collected data from the sensors on the wall, logged sensor activation, and played sounds in response to sensor activations. We selected an Arduino Mega to serve as the main microcontroller for the system. An Adafruit micro-SD card breakout board, DS3231 real-time clock module (RTC), and DY-SV5W media player board were soldered onto an Arduino Mega proto-shield which allowed the additional components to connect to the Arduino Mega host controller via 0.1” header pins, all of which can be seen in Figure 8. For each hole on the wall, we installed a VL53L0X time of flight (ToF) sensor alongside a Digispark Attiny85 microcontroller. We designed the system such that anytime an elephant put their trunk through the enrichment wall and across the sensor, the microcontroller would register the change in proximity and send a positive signal back to the Mega to document which hole the elephant interacted with. This was done to reduce potential capacitive interference with the analog signal coming from the ToF sensors as they were relayed across lengths of cable between 15 and 30 feet. Each time a sensor was activated, the current time and date (provided by the RTC) was logged along with the index of the hole which was activated within a.csv file stored on the micro SD card. The ToF sensors were monitored at 200 Hz but the data was only recorded to the.csv file whenever a state change occurred (activated to not activated or not activated to activated).
Figure 8:
After the sensor activation was logged, the Arduino Mega sent a serial command to the media player which selected and played a tone from a list of.mp3 files stored on the onboard micro SD corresponding to the specific insert interacted with. The tones were of the following selected frequencies: 61.74 Hz, 55.00 Hz, 49.00 Hz, and 43.65 Hz. We selected these based on the results from French et al. showing that the elephants prefer frequencies around 60 Hz [11]. The custom-built outdoor speaker (pictured in Figure 9) consisted of a Pyle PMRA402 marine amplifier, Sony XS-MP1611 marine drivers, and a 12v 10a DC waterproof power supply in a resilient, rubber enclosure. The speaker was then connected to the media player and played one of the four corresponding tones to provide the acoustic response to the elephant.
Figure 9:
4.4.3 Prototype Installation and Development.
To ensure the safety and robustness of our design, we followed an iterative deployment process for prototype testing. This involved phased installations, beginning with a preliminary version that did not include electrical components. This initial stage allowed us to assess the system’s durability in a controlled manner before introducing the full prototype to the elephants. The iterative approach facilitated continuous improvement, enabling us to refine the design based on the gathered data.
4.4.4 Pilot Testing.
The initial phase of prototype evaluation involved pilot testing to assess elephant behavior towards the installation and the insert’s durability. Specifically, the pilot tests aimed to determine whether elephants would purposefully avoid the hole containing the insert, the propensity of the elephants to destroy the insert, and the insert’s ability to withstand the forces exerted by an elephant during interaction. We installed the prototype prior to the zoo’s opening while the elephants were housed in their indoor barn. The first iteration utilized high-impact polystyrene (HIPS) vacuum-formed bevels at both ends of the PVC pipe [Figure 10]. One bevel was permanently affixed with epoxy, while the other was fastened using polycarbonate nuts and bolts for facilitating on-site installation. The insert was strategically placed in a hole closest to the most readily accessible food source. Following installation, we began observations as soon as the elephants were released into the habitat. Our observations focused on their behavior towards the insert and the prototype’s overall structural performance, which were also documented on camera. During this first attempt, the insert had a near-immediate breakage of the HIPS bevels under pressure from an elephant’s trunk. As a result, the second pilot test instead implemented three right-angled metal brackets, custom-bent to match the bevel angle, and foam insulation for securing the insert within the wall [Figure 10]. This iteration was initially successful with elephant interaction and therefore remained installed for a period of three days.
Figure 10:
4.4.5 Full Installation.
The full installation phase involved the installation of the complete prototype, incorporating both the physical structure and the electrical components. Four inserts were strategically positioned within the wall, occupying all of the holes on the second row from the bottom, chosen due to their close proximity to the feeders behind the wall and for ease of installation. Once deployed, this version of the system experienced failures from false sensor activations. The time-of-flight sensors proved susceptible to dirt accumulation, triggering the speaker system without an elephant trunk being present. Additionally, the elephants’ behavior, involving pushing their trunks against the inserts, resulted in bending of the metal brackets.
Figure 11:
Figure 12:
We implemented a revised prototype to address these issues and ensure the integrity of both the system and the elephants’ well-being. This iteration prioritized the separation of the elephants’ physical interaction from the sensitive electronic components. PEX tubing replaced the original design, with ratchet straps securing the tubing to the back of the wall. Similar to the previous setup, the ToF sensors and Attiny85 microcontrollers were housed within coupling pieces positioned above the four designated holes. This revised prototype was successful and remained accessible to the elephants for ten days, facilitating continued data collection and behavioral observations.
4.5 Data Collection
4.5.1 Experiment Procedures.
Before the system was installed, ten days of continuous video data was collected of the elephants interacting with the existing enrichment wall in the habitat to serve as a control session. Once the system was installed, the system was to be running continuously for ten days and available to all three elephants anytime they were within the habitat. The main goals of this test were to determine how successful the device was at maintaining structure against wear, track usage of the enrichment wall, and evaluate if there would be an increase in engagement with the audio augmentation compared to the wall before any alteration. To compare engagement, ten days of continuous video data were collected while the device was installed, and then compared to the previous control session of ten days of video data taken before we implemented any changes to the enrichment wall. We relied on these videos to determine the number of minutes spent at the wall, the number of instances an elephant approached, and which elephant was engaging with the wall. Sensor data was used to determine which specific holes were being interacted with, and to compare accuracy of device usage with the video.
It is important to note the extraneous variables that were not able to be adjusted regarding the experiment period and enrichment wall: the keepers had independent control over the placement of food, how often the food was stocked, what type of food, and the amount of food the elephants were given.
4.5.2 Video Annotations.
As the data from our prototype would only tell us when a specific hole was activated, we wanted a method of validating this data while also giving us more insight into which elephant was interacting with the wall. To do this, we installed a Google Nest Camera directly outside of the habitat, pointed at the front side of the enrichment wall. To compare wall usage between elephants before and after the installation, two sets of 10-day video recordings were downloaded in 15-minute increments off of the camera, one taken before installation and one taken after. We then used Adobe Premiere Pro, a professional video editing software, to label the videos [1]. Labeling was done by creating a caption with the name of the elephant that lasted from the moment they approached the wall to the moment they left. The annotations were saved to separate.csv files, each named with the initial timestamp of the corresponding video. This allowed us to identify which elephant was approaching, and record the number of approaches at the wall as well as the retention time for every approach. After annotations were completed, all of the.csv files were appended to a dataframe using the pandas python library [24]. We separated the.csv files based on whether they were from before or after the installation. Duration times were calculated from the start and end times of each annotation. Then, we calculated the total number of annotations and total duration time for each elephant. The total number of annotations represented the number of instances an elephant approached the wall and the duration time was the total time spent standing at the wall. This not only included interacting with the holes but also eating the hay they had just reached for. From these annotations, we learned the difference in usage patterns between the three elephants as well as the change in usage patterns, if any, caused by the installation of our prototype.
4.5.3 System Reliability Analysis.
The sensor dataset included timestamps and binary values for each of the four holes, recorded whenever any of the values changed, telling us when a sensor was activated and deactivated. To visualize the sensor data, we resampled it to 4 Hz. Since the timestamps were collected at seconds precision (without milliseconds), there were multiple readings with the same timestamp. Duplicate values were removed before resampling. Then, the dataset was split based on each hole and plotted to view actuation periods. The scipy find peaks function (with the height parameter set to 1) was used to calculate the number of times each hole was interacted with [7].
4.6 Dashboard
To display the data from our device to the elephant keepers in a clear and helpful manner, we created two designs for a data dashboard: a functional dashboard and a prototype dashboard using Figma, a collaborative interface design tool [9].
4.6.1 User Interviews.
To understand what features would best support the keepers in their care of the elephants, we conducted semi-structured interviews with four of the elephant keepers at the zoo. Our goals were to learn what data is most important to them to see in our dashboard and the best way to display that data. To gain this information, we asked the following questions:
(1)
What kinds of data do you normally collect on the elephants?
(a)
Where do you record this information?
(2)
How does the data you collect determine your evaluation of the elephant’s welfare?
(3)
What data do you wish you could keep track of that you can’t currently?
(4)
With regard to their interaction with the wall, what data are you interested in seeing?
(5)
What timeframes would you like to see the data in?
(6)
Would you like to see comparisons between elephants?
(a)
If so, what specific data points would you like to see compared between them?
(7)
Other than interactions with the wall, what kind of info would you like to be able to see in the app, if any?
To organize and analyze our insights from these interviews, we conducted one round of qualitative coding by forming an affinity diagram of the notes that were transcribed during each of the interviews. These consolidated findings were then used to inform our designs of the dashboard.
4.6.2 Functional Dashboard.
Using an open-source Python library called Streamlit, we designed a functional dashboard to visualize the data recorded from our video annotations [30]. Streamlit allowed us to view graphs in different time frames and gave the keepers a closer idea of what a fully functional online platform would look like for visualizing the data of the enrichment wall.
In the dashboard, the user has the capability of choosing between three timeframes: 1 day, 1 week, or 10 days, which is the longest amount of time we consecutively collected data. Additionally, the user can select to view data for all three elephants, or they can choose to view individual data for one of the three elephants. In any of these views, a corresponding bar graph or line chart is displayed to the user. These pages can be viewed in Appendix A.
4.6.3 Figma Prototype.
We used Figma to build out a simulated interactive prototype of the proposed designed system we envisioned elephant keepers would interact with on a day-to-day basis to accompany the enrichment wall device. Visual design components and assets such as font choice, primary and secondary color usage, and branding were derived from the zoo’s publicly available branding guidelines and existing interface design across their website and social media channels.
The interface consists of two primary tabs: the tab for collective herd engagement with the wall device, and a tab that shows more of each of the elephants’ independent statistics and data. The entire interface has a time select feature that allows keepers to view data selected from today, the previous 7 days, any previous day, or a custom date range. The main screens are included in Appendix B.
Primary features of the Wall tab:
•
A data visualization of the enrichment wall displaying the number of trunk insertions for each specific hole for the given date range with a higher color saturation value associated with holes with a higher engagement number.
•
A picture upload feature for keepers to document the enrichment setup behind the wall each day. This allows keepers to keep track of the data associated with the feeders placed behind the wall and deduce what enrichment products the elephants might be engaging with more and therefore prefer.
•
A bar chart displaying the total time each elephant has spent engaging with the wall for the current date or date range selected.
•
An engagement over time data visualization. During a single-day selection, this will be presented as a Gantt chart to show the number of approaches over the day, and time spent engaging with the wall per session. During a date range selection, this will be presented as a line graph comparing each elephant’s engagement time with the wall over the selected dates.
Primary features of the Eles tab:
•
A section for each herd member with at-a-glance statistics representative of the selected date range such as total time at the wall, preferred insert, if any, and favorite enrichment feature derived from collected data.
•
Each section is a clickable element that opens a more detailed full-screen presentation of that specific elephant’s data including further detailed statistics, and a graph presenting engagement over time.
4.6.4 Usability Testing.
To evaluate the designed interface, our team outlined a usability testing guideline to be tested with four elephant keepers at the zoo. The conducted sessions involved each keeper being presented with a series of tasks to complete through the prototyped Figma interface while being asked to say their thoughts aloud as they navigated.
The tasks presented were as follows:
(1)
View today’s data and find which hole is being used the most that day
(2)
For today’s data, compare the time spent at the wall between elephants and decide which elephant used the wall the most that day
(3)
Upload a photo of the feeder placement on the wall for that day
(4)
Navigate to Elephant 1’s information
(5)
View Elephant 1’s data for today and determine his favorite hole to interact with
Following the task session, the participant was asked a series of Likert scale questions targeted at evaluating the ease and usefulness of the tasks and features presented. The number of questions asked to the researcher was documented for each task in addition to whether they were able to effectively complete the task. General feedback and notes were recorded both during the think-aloud portion and open discussion following the session.
5 Results
5.1 Research and Ideation
5.1.1 Qualitative Research.
Insights gained from the initial qualitative research of our contextual inquiry, semi-structured interviews, and observation informed the development of the method of instrumenting the enrichment wall and was further narrowed by a sequential meeting with the elephant keepers.
With regard to enrichment devices and activities, the keepers noted that their elephants are motivated to engage with enrichment devices when food rewards are incorporated. The keepers mentioned that these moderately geriatric elephants (ages 36-41), do not show interest in interaction unless food is incorporated as a reward. Their motivation for food was evident during our observation sessions, where we continuously saw the elephants moving between different food sources throughout their habitat. Zookeepers at other zoos, however, mentioned that the elephants under their care displayed play behaviors with a variety of enrichment devices and engagements, even if a food reward was not included.
5.2 Fabrication and Installation
5.2.1 Pilot Testing.
During the initial pilot test with the first prototype, the insert broke after 1.5 hours of being in the habitat. As one of the elephants inserted their trunk into the hole we had installed in, she tried to reach hay far away from the wall and the pressure of her trunk snapped the HIPS bevel in half. We were informed by the keepers that the HIPS bevel being used as a housing mechanism for the front of the wall was like “handing the elephants a paper plate” and we needed to come up with a stronger design for the elephants to directly put pressure on and interact with. The second round of testing saw the insert remain structurally sound through its 3 day testing period.
5.2.2 Full Installation.
The interactive installation was originally scheduled to be operable for 10 days but ultimately this period was cut short to 7 days and 10 hours due to a hardware failure in the system. Throughout the study period, there were several occurrences of false activations of the speaker without an elephant present (due to hay blowing in front of the sensor). Anytime the speaker was outputting continuously, while an elephant was not present, one of the keepers would clean the sensors to dislodge any stuck pieces of hay. At the 7-day and 10-hour mark, however, two of the holes stopped working accurately due to one of the connectors into the electronics box failing. Shortly after that, a solder joint broke on one of the wires which supplied power to the sensors. These hardware failures resulted in three of the four sensors providing false readings, which caused false positive elephant trunk detection and false positive speaker activation. These failures occurred in an area where a lot of movement happened daily by the keepers, so we cannot state whether this was caused by an elephant or a human. However, throughout the study, the keepers said they didn’t notice any of the elephants deliberately trying to break our system. Even with the hardware malfunctions, we were successful in collecting the sensor data from the onboard microSD card.
5.3 Data Collection and Analysis
5.3.1 Elephant Enrichment Engagement Measured from Video Annotations.
The dataframes were filtered out to only include 7 days and 10 hours of data (from 9 am on day 1 to 7 pm on day 8) to remove any data that was collected after the hardware malfunctions. The duration times and number of approaches at the wall before and after the installation can be seen in Tables 3 and 4. The average duration time increased by 71.3% from 4 hours, 10 minutes, and 7 seconds to 7 hours, 8 minutes, and 30 seconds. However, this increase was not statistically significant (p=0.0561, t-stat = -4.0419, alpha=0.05). The average frequency of usage (measured by the number of approaches at the wall) increased significantly by 176%, from 46 to 127 (p=0.0151, t-stat=-8.0465, alpha = 0.05).
Table 3:
| Elephant | Time Duration at Wall (pre) (h:m:s) | Time Duration at Wall (post) (h:m:s) |
|---|---|---|
| 1 | 04:29:31 | 06:41:26 |
| 2 | 06:43:18 | 11:09:55 |
| 3 | 01:17:32 | 03:34:10 |
| Total | 12:30:21 | 21:25:31 |
Table 4:
| Elephant | Approaches at Wall (pre) (n) | Approaches at Wall (post) (n) |
|---|---|---|
| 1 | 47 | 116 |
| 2 | 79 | 180 |
| 3 | 12 | 85 |
| Total | 138 | 381 |
5.3.2 System Reliability Analysis.
Using the video annotation data, we were able to filter out any sections of the sensor data where the sensor was likely activated by a piece of hay since we knew what times an elephant was at the wall or not. To account for any offset between our video annotation times and the sensor data timestamps, we added a buffer of 10 seconds to the start and end times of the video annotations. Even with filtering out the times when an elephant was not there, we knew there were still false actuations in the sensor dataset, as there were rows within the sensor data dataframe where more than one hole was activated high at the same time. To calculate the false positive rate as shown in Table 5, we included the following as false actuations: the number of times at least one sensor was activated in the time periods we did not have video annotations for and the number of times more than one sensor was activated in the times there was an elephant at the wall. The rest of the times that only one sensor was activated were counted as true actuations. This gave us a false positive rate of 11.858%, meaning our system was 88.14% accurate in detecting when an elephant was interacting with one of the four holes.
Table 5:
| False Actuations | True Actuations | Total Actuations | False Positive Rate |
| 464 | 3449 | 3913 | 11.858% |
After resampling, the sensor data was plotted to visualize hourly data for an individual hole. A plot visualizing the data for hole 1 for one hour in the afternoon (13:00-14:00) on the 7th day can be seen in Figure 13. Additionally, the total number of interactions per hole can be viewed in Table 6. While the numbers in the table correlate with Hole 1 being the most interacted with hole, these values contain errors due their inclusion of instances where two sensors may have been activated at the same time.
Figure 13:
Table 6:
| Hole | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| Number of Interactions | 975 | 233 | 214 | 216 |
5.4 Dashboard
5.4.1 User Interviews.
From the user interviews informing the dashboard design, we learned that it is hard for the keepers to know what the elephants are doing at all times of the day. When asked what kind of data they would like to see regarding the wall, they expressed that they would like to know the retention times at the wall, who is using the wall and how often, whether hierarchy or personality affects the usage of the wall, and what feeders and openings are interacted with the most. Last, they expressed the importance of the data being collected, as it could justify the need for a second enrichment wall since some of the elephants get to use the wall more than others.
Table 7:
| Task Number | Task | Easiness Score | Usefulness Score |
|---|---|---|---|
| 1 | View today’s data | 4.5 | 4.75 |
| for all elephants | |||
| 2 | Upload a photo | 4.75 | 5 |
| of feeder setup | |||
| 3 | View today’s data | 4.25 | 5 |
| for Elephant 1 |
5.4.2 Usability Testing.
Results from the usability testing session were overall positive. Ratings for each task based on easiness and usefulness (1 being worst, 5 being best) are shown in Table 7. Keepers especially found the photo upload feature very useful to correlate the data with what enrichment devices/food containers were placed behind the wall that day. Half of the keeper participants attempted to navigate to the individual elephant data via an infographic rather than through tabs provided at the bottom of the screen, causing a lower easiness score for task 3. From further observation, it was apparent that the tabs shown at the bottom screen of the interface were not as clearly understood to be buttons. Furthermore, 2/4 of the users wanted to click on the red circles that represent the holes on the enrichment wall to view a further breakdown of that data and expressed that they looked like clickable buttons. Additional critical feedback is that the general interaction over time graph was difficult to read, and the graph titling was a bit unclear across the different graphs on the wall page.
6 Discussion
The findings from this study show that all three elephants spent more time interacting with the augmented enrichment wall than before our device was installed. This is especially interesting knowing that the keepers had put the elephants on a diet at the same time as our installation, meaning they were being fed less hay in the enrichment wall feeders than they were before installation. Elephant 3 was typically known to not be at the wall often, as she is at the bottom of the hierarchy in the herd, and yet her time duration spent at the wall almost tripled after installation. Additionally, the keepers commented that two of the three elephants were observed interacting with the instrumented enrichment holes even when there was no hay behind the wall. When only two of the holes were operational, we observed Elephant 1 moving deliberately between the holes, activating the sounds when there was no hay. This suggests a possibility that the elephants are curious about the sounds coming from the speaker. However, further studies would have to be conducted to determine whether they were truly interested in interacting with our system for sound enrichment, or if they approached the wall more often for alternative reasons. These alternative reasons could have been because they knew that zoo visitors would pay attention to them when they activated the sounds, or because of the false activation times acting as an audio beacon bringing curiosity to the elephants. The audio enrichment from the speaker was also helpful for the keepers, providing them with auditory cues that could tell them that an elephant was using the wall.
When it comes to designing an interactive enrichment device for elephants, we reinforced the findings from French’s works [10, 11] that even if you think you have made your system as durable as possible, an elephant can still find a way to break it. However, unlike in French’s works, our elephants showed no deliberate intent in breaking our device; it merely broke from the sheer amount of force the elephants used when pushing up against the wall. We learned that the best way to avoid destruction was to put the device on the back side of the wall so they couldn’t directly push against it. We also learned that utilizing the least amount of hardware is ideal. It was important to iteratively prototype the system to find the most durable design, but also to allow us to develop the sensing mechanisms before exposing the elephants to actuation. Additionally, designing at a scale for elephants is challenging and expensive. What we ended up utilizing happened to be simplistic items that already existed (PEX tubing, PVC conduit boxes, and ratchet straps) instead of custom-designed pieces. This also helped us overcome the limitations of designing for that specific wall, since each hole had slightly different measurements. A solution with even less hardware, such as a camera system utilizing computer vision, may prove to be even more successful.
7 Future Work
The course of this project provides a basis for a number of future work efforts both expanding upon the designed system and creating new ones for more detailed quality of life analysis for zoo-housed elephants. As our system only instrumented four of fourteen holes, there is a clear path to instrumenting all holes for a total comprehensive analysis of engagement with the existing enrichment wall. As there were issues with elephants breaking the device due to contact and some faulty electronic issues, computer vision might be used in future work to resolve these, as it would remove any potential direct contact with a device and significantly reduce infrastructure. To more easily determine which elephant was interacting with the wall, RFID and/or GPS tags could be fitted onto the elephant’s body, and tracked such that we not only know which elephant is at the wall when interacting with it but also determine each elephant’s use of the wider habitat. From learning their general space usage throughout the day, we could measure how their usage of the enrichment wall compares to other types of enrichment in their habitat, such as hanging feeders or barrels, to better understand their enrichment preferences.
Beyond the scope of enrichment through the wall, keepers also expressed the desire to know how much food and water each elephant was consuming each day. A system such as ours could allow the keepers to track consumption data and be aware of any deviation or aid in diet tracking. This might be done by equipping water troughs and food containers with load cells. Ultimately, the elephant keepers were interested in the data from our device and eager to continue working with us for future projects.
8 Conclusion
There have been many different types of enrichment provided to elephants in zoos, and researchers have explored designing interactive enrichment systems for elephants to give them more control over their habitat. However, to our knowledge, our project is the first to design and deploy an interactive enrichment device incorporating electronics for elephants in a longitudinal study and successfully collect data. Our system utilized sensors attached to an existing enrichment feeder wall to monitor usage as well as provide sound enrichment, increasing the time spent at the wall by 3 hours, 28 minutes, 23 seconds (71.3%) and the frequency of approaching the wall by 81 instances (176%). This showed that the elephants responded positively to the sound enrichment. Our hardware performed with an accuracy of 88.14% for 7 days and 10 hours. Future work involves utilizing camera footage of the elephants using the wall to apply computer vision techniques to avoid future hardware issues or long-term wear and tear of the current physical system.
Acknowledgments
We would like to thank the Zoo Atlanta keepers and elephants who participated in this study. We would also like to thank Fiona French, Tim Trent, Noah Posner, Tripp Edwards, and Chris Simon for their system design and prototyping advice. This work was also completed thanks to the resources available in the IPaT Prototyping and Craft Labs and the Georgia Tech Design Shop.
A Functional Dashboard
Figure 14:
B Figma Prototype
Figure 15:
References
[1]
Adobe. 2024. Adobe Premiere Pro. https://www.adobe.com/products/premierepro.html.
[2]
Elephant Managers Association. 2024. The Elephant Managers Association - Home. https://elephantmanagers.com/
[3]
AZA. 2024. Association of Zoos and Aquariums (AZA. https://www.aza.org/
[4]
Ceara Byrne, Sabrina Karl, Dana Vilker, Sebastian Meller, Brendan Thompson, and Sarah Webber. 2020. Monitoring the Welfare of Bears in Captivity. In Proceedings of the Seventh International Conference on Animal-Computer Interaction (Milton Keynes United Kingdom, 2020-11-10). ACM, 1–6.
[5]
Ceara Byrne, Jay Zuerndorfer, Larry Freil, Xiaochuang Han, Andrew Sirolly, Scott Cilliland, Thad Starner, and Melody Jackson. 2018. Predicting the Suitability of Service Animals Using Instrumented Dog Toys. 1, 4 (2018), 127:1–127:20.
[6]
Jon Coe and Julia Hoy. 2020. Choice, Control and Computers: Empowering Wildlife in Human Care. Multimodal Technologies and Interaction 4, 4 (Dec. 2020), 92. Number: 4 Publisher: Multidisciplinary Digital Publishing Institute.
[7]
SciPy developers. 2024. SciPy: open-source scientific computing library for Python. https://scipy.org/.
[8]
Alessia Diana, Marina Salas, Zjef Pereboom, Michael Mendl, and Tomas Norton. 2021. A Systematic Review of the Use of Technology to Monitor Welfare in Zoo Animals: Is There Space for Improvement?11, 11 (2021), 3048. Number: 11 Publisher: Multidisciplinary Digital Publishing Institute.
[9]
Figma. 2024. Figma: A design platform for teams. https://www.figma.com/.
[10]
Fiona French, Clara Mancini, and Helen Sharp. 2018. High tech cognitive and acoustic enrichment for captive elephants. 300 (2018), 173–183.
[11]
Fiona French, Clara Mancini, and Helen Sharp. 2020. More Than Human Aesthetics: Interactive Enrichment for Elephants. In Proceedings of the 2020 ACM Designing Interactive Systems Conference (New York, NY, USA, 2020-07-03) (DIS ’20). Association for Computing Machinery, 1661–1672.
[12]
Brian J. Greco, Cheryl L. Meehan, Jennifer L. Heinsius, and Joy A. Mench. 2017. Why pace? The influence of social, housing, management, life history, and demographic characteristics on locomotor stereotypy in zoo elephants. Applied Animal Behaviour Science 194 (Sept. 2017), 104–111.
[13]
Juan Haladjian, Johannes Haug, Stefan Nüske, and Bernd Bruegge. 2018. A Wearable Sensor System for Lameness Detection in Dairy Cattle. 2, 2 (2018), 27. Number: 2 Publisher: Multidisciplinary Digital Publishing Institute.
[14]
Ilyena Hirskyj-Douglas and Vilma Kankaanpää. 2022. Do Monkeys Want Audio or Visual Stimuli? Interactive Computers for Choice with White-Faced Sakis in Zoos. In Designing Interactive Systems Conference (Virtual Event Australia, 2022-06-13). ACM, 1497–1511.
[15]
Vilma Kankaanpää, Fay E Clark, and Ilyena Hirskyj-Douglas. 2024. LemurLounge: Lemurs’ Individual-Level, Group, and Cross-Species Use of an Interactive Audio Device in Zoos. In Proceedings of the CHI Conference on Human Factors in Computing Systems (Honolulu HI USA, 2024-05-11). ACM, 1–14.
[16]
Vilma Kankaanpää and Ilyena Hirskyj-Douglas. 2023. Prototyping with Monkeys: Uncovering What Buttons for Monkeys Look Like. In Proceedings of the Seventeenth International Conference on Tangible, Embedded, and Embodied Interaction (Warsaw Poland, 2023-02-26). ACM, 1–13.
[17]
Rébecca Kleinberger, Anne H. K. Harrington, Lydia Yu, Akito Van Troyer, David Su, Janet M. Baker, and Gabriel Miller. 2020. Interspecies Interactions Mediated by Technology: An Avian Case Study at the Zoo. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (Honolulu HI USA, 2020-04-21). ACM, 1–12.
[18]
Clara Mancini. 2011. Animal-computer interaction: a manifesto. interactions 18, 4 (July 2011), 69–73.
[19]
Hal Markowitz. 1982. Behavioral Enrichment in the Zoo. Van Nostrand Reinhold.
[20]
Hal Markowitz, Cheryl Aday, and Anita Gavazzi. 1995. Effectiveness of acoustic “prey”: Environmental enrichment for a captive African leopard (Panthera pardus). Zoo Biology 14, 4 (1995), 371–379. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/zoo.1430140408.
[21]
Christopher Flynn Martin, Dora Biro, and Tetsuro Matsuzawa. 2011. Chimpanzees’ use of conspecific cues in matching-to-sample tasks: public information use in a fully automated testing environment. 14, 6 (2011), 893–902.
[22]
Christopher Flynn Martin, Akiho Muramatsu, and Tetsuro Matsuzawa. 2022. Apex and ApeTouch: Development of a Portable Touchscreen System and Software for Primates at Zoos. 12, 13 (2022), 1660. Number: 13 Publisher: Multidisciplinary Digital Publishing Institute.
[23]
Cheryl L. Meehan, Jennifer N. Hogan, Mary K. Bonaparte-Saller, and Joy A. Mench. 2016. Housing and Social Environments of African (Loxodonta africana) and Asian (Elephas maximus) Elephants in North American Zoos. PLOS ONE 11, 7 (July 2016), e0146703. Publisher: Public Library of Science.
[24]
The pandas development team. 2023. pandas-dev/pandas: Pandas.
[25]
Bonnie M. Perdue, Andrea W. Clay, Diann E. Gaalema, Terry L. Maple, and Tara S. Stoinski. 2012. Technology at the Zoo: The Influence of a Touchscreen Computer on Orangutans and Zoo Visitors. 31, 1 (2012), 27–39. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/zoo.20378.
[26]
Charles Ramey, Arianna Mastali, Cole Anderson, William Stull, Thad Starner, and Melody Jackson. 2023. WAG’D: Towards a Wearable Activity and Gait Detection Monitor for Sled Dogs. In Proceedings of the Ninth International Conference on Animal-Computer Interaction (New York, NY, USA, 2023-03-29) (ACI ’22). Association for Computing Machinery, 1–13.
[27]
Capturing Reality. 2024. RealityCapture: Photogrammetry Software. https://www.capturingreality.com/.
[28]
Nancy L. Scott and Chase A. LaDue. 2019. The behavioral effects of exhibit size versus complexity in African elephants: A potential solution for smaller spaces. Zoo Biology 38, 5 (2019), 448–457. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/zoo.21506.
[29]
Joseph Soltis. 2010. Vocal communication in African Elephants (Loxodonta africana). Zoo Biology 29, 2 (2010), 192–209. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/zoo.20251.
[30]
Streamlit. 2024. Streamlit: The fastest way to build custom data apps for everyone. https://streamlit.io/.
[31]
Sarah Webber, Marcus Carter, Wally Smith, and Frank Vetere. 2020. Co-Designing with Orangutans: Enhancing the Design of Enrichment for Animals. In Proceedings of the 2020 ACM Designing Interactive Systems Conference (Eindhoven Netherlands, 2020-07-03). ACM, 1713–1725.
[32]
E. Williams, C. L. Chadwick, L. Yon, and L. Asher. 2018. A review of current indicators of welfare in captive elephants (Loxodonta africana and Elephas maximus). Animal Welfare 27, 3 (Aug. 2018), 235–249.
[33]
Yating Wu, Harpreet Sareen, and Gabriel A. Miller. 2021. Soundyssey: Hybrid Enrichment System for Elephants in Managed Care. In Proceedings of the Fifteenth International Conference on Tangible, Embedded, and Embodied Interaction (Salzburg Austria, 2021-02-14). ACM, 1–6.
[34]
Lisa Yon, Ellen Williams, Naomi D. Harvey, and Lucy Asher. 2019. Development of a behavioural welfare assessment tool for routine use with captive elephants. PLOS ONE 14, 2 (Feb. 2019), e0210783. Publisher: Public Library of Science.
Index Terms
- Play That Trunky Music: Development of an Auditory Enrichment Device for Elephants in Zoos
Recommendations
Music perception and cognition following bilateral lesions of auditory cortex
We present experimental and anatomical data from a case study of impaired auditory perception following bilateral hemispheric strokes. To consider the cortical representation of sensory, perceptual, and cognitive functions mediating tonal information ...
Soundyssey: Hybrid Enrichment System for Elephants in Managed Care
TEI '21: Proceedings of the Fifteenth International Conference on Tangible, Embedded, and Embodied InteractionAnimals in managed care ideally are provided with environmental stimuli for their psychological and physiological well-being. Most commonly, food-based enrichment methods are used to mimic wild conditions, allowing animals to search and forage. Other ...
Auditory imagery modulates frequency-specific areas in the human auditory cortex
Neural responses in early sensory areas are influenced by top-down processing. In the visual system, early visual areas have been shown to actively participate in top-down processing based on their topographical properties. Although it has been ...
Comments
Information & Contributors
Information
Published In
December 2024
180 pages
ISBN:9798400711756
DOI:10.1145/3702336
Copyright © 2024 Copyright held by the owner/author(s).
This work is licensed under a Creative Commons Attribution International 4.0 License.
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Published: 02 December 2024
Check for updates
Author Tags
Qualifiers
- Research-article
Conference
ACI 2024
ACI 2024: The International Conference on Animal-Computer Interaction
December 2 - 5, 2024
Glasgow, United Kingdom
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 547Total Downloads
- Downloads (Last 12 months)547
- Downloads (Last 6 weeks)151
Reflects downloads up to 01 Mar 2025
Other Metrics
Citations
View Options
View options
View or Download as a PDF file.
PDFeReader
View online with eReader.
eReaderHTML Format
View this article in HTML Format.
HTML FormatLogin options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in