Duration
9 Months
Year
2023-2024
My Role
UX Design
Fabrication
Physical Prototyping
Team
Ben Mayo
Arianna Mastali
Charles Ramey
Dr. Melody Jackson
Bringing stimulating enrichment and data tracking to captive elephants
As advancing technology has allowed for new and inventive methods of human-computer interaction, so too can we use technology to create new and inventive methods of animal-computer interaction. With any animal in captivity, sufficient physical and cognitive engagement is necessary to maintain healthy welfare. In the case of animals as large and intelligent as elephants, this becomes all the more difficult.
We designed, fabricated, and tested an acoustic enrichment device at Zoo Atlanta that lets the elephants play sounds as they insert their trunks into slots in an existing feeding wall in their habitat. Using proximity sensors and microcontrollers, we enabled the elephants to play different sounds across each slot, and tracked their interaction over time to compare with their previous interaction with the feeding enrichment wall. With the acoustic device installed, we saw the elephants spend nearly twice as much time engaging with the wall as before, and developed a conceptual prototype for a digital dashboard that can present this data in an effective way to the handlers.
Our users are non-English speaking, 7,000 pound packages of potential destructive force with incredibly dexterous trunks. To better understand how we can design for these animals, we delved into a variety of different avenues for research. Firstly, a literature review of natural elephant behaviors to discover elephants spend approximately 18 hours a day foraging for food, have keen senses of smell and hearing, especially good tactile abilities with their trunk, and poor eyesight. Pursuing enrichment, we interviewed zoo staff elephant handlers and Fiona French, a professor who has researched elephant interaction with buttons, foot pedals, and trunk proximity sensors. Entering the zoo space, we additionally read through the American Zoos and Aquariums(AZA) standard care procedures for captive elephants, which Zoo Atlanta follows, and joined the Elephant Managers Association (EMA). Through the EMA we could access their enrichment product database which included community designed and tested toys.
Following our interviews with the Zoo Atlanta handlers, we established foundational knowledge on what a successful enrichment device for these specific elephants would look like. The three elephants (Msholo (M) - 37, Kelley (F) - 40, Tara (F) - 40) are considered geriatric, and from enrichment discussions, are very heavily food motivated. We were told: "if there's not food in or on it, they're not interested."
We set about a brainstorm session especially keeping in mind existing enrichment protocols, with the goal to create a device that would add a level of complexity to their existing environment, allow some method of data tracking, and be easily trainable with food. Any concept took note of the elephants prominent senses, especially hearing, scent, and touch. What came from that was a set of three, further developed concepts sketched below: a capacitive touch wind chime, an augmented enrichment wall, and a personal elephant heater for cold, winter days.
Presenting the concepts to the zoo, we pursued the enrichment wall due to its leverage in using an existing piece of habitat infrastructure that they regularly engage with (their enrichment feeding wall), and the flexibility in outputs, as we can use a trunk trigger to do a multitude of things. However, for primary research with a controlled variable, we pursued acoustic enrichment, as the elephants do not currently have that as part of their program.
Building a device into an existing piece of infrastructure means that we need to document it as accurately as possible in order to design around it. We used photogrammetry to capture sections of the wall that were brought into a 3D modeling software (seen here) and used as the basis for the trunk-detection inserts to ensure an as accurate as possible fit. Using the same model, we CNC routed a section of the wall from foam to test fit our prototypes during development without the need to travel to the zoo.
With the strength of the elephant in mind, and the constraint that we cannot drill into or permanently alter the wall in any way, we designed an insert that can host a time of flight sensor used to determine the presence of a trunk. An 8" diameter PVC pipe section forms the structural core, while flanges on each end act as plugs to keep it snug in the hole. A cap on the end of one flange covers the sensor and electronics inside so that there's no risk of the elephants accessing them.
The first step of fabrication was to create a mold onto which we could vacuum form sheets of polystyrene to form the flange pieces. Using the geometry from the wall scan, we made a 3D model of the inverse of the wall's inner chamfer, then CNC milled the form from MDF. Each sheet of polystyrene was heated, then vacuumed over the form to create the flange. The 8" diameter PVC was cut into approximately 15" lengths, acrylic sheets were cut to form the cap that encloses the electronics on one end, and once put in place, the plug end with electronics and acrylic cap were adhered in place. The flange on the other end is left removable thanks to removable nylon bolts to allow for installation.
Initial onsite testing of the first prototype ended quickly when Kelley, the 41 year old female, pushed her trunk into the hole containing the insert, which quickly broke the front polystyrene flange holding the insert in place and pushed it out the back of the wall. Understanding better that whatever holds the insert in from the front of the wall will need to stand up to this force, we successfully tested an improved prototype that relied on removable, pronged metal brackets bolted into the PVC insert body to replace that front polystyrene flange.
The full scale test resulted in a failure after 3 days when the metal brackets holding one of the inserts was bent and pushed out due to the pressure of the elephants inserting their trunks again, in addition to dirt accumulation on the proximity sensor causing false positives.
Learning from the difficulty that arose from an insert that was built to combat the elephant's strength, we concluded the best design is one that avoids any close contact at all. While maintaining the use of proximity sensors, the new design is constructed from lengths of PEX tubing and electrical conduit boxes. Each conduit box houses a small microcontroller and time of flight sensor to determine proximity of objects passing in front. The PEX tubing allows wires reaching each of the nodes to connect to the main microcontroller, which logs which hole was triggered by an interaction to an onboard SD card, and sends an audio signal out to the speaker to play the associated tone.
Below you can watch a clip of Msholo standing with his trunk in one of the holes that triggers a sound. I've placed a high pass filter on the audio to clip out the nearby waterfall. If you turn up the volume you can hear the low tone coming out of the speaker.
While our device collects data, we need a way to effectively share the insights with the handlers, and so began design on a dashboard concept. Data transparency and insights are the primary goal, so our dashboard's homepage hosts essential tracking data visualizations for easy, on sight comparison. Secondly, individual elephant statistics for comparison let handlers easily evaluate the share of engagement between each member of the herd. Social hierarchy also shows up here, as Msholo, the bull, spends the most time at the wall as the dominant male, while Tara regularly spent the least time at the wall, being the lowest on the totem pole.
With primary features developed and vetted with the handlers, we continued to develop a testable prototype complete with design guidelines informed by Zoo Atlanta's existing branding. As form follows function, the interface is designed with readability of information in mind to best present interaction data to the handlers. Brand colors are used for data highlights, such as on the primary wall map data visualization, where color intensity is used to indicate greater use for better at-a-glance reading.
Handlers have the ability to create date ranges to compare interaction over time, and discern any noticeable changes in trend. Significant changes in engagement can help to signify an underlying health problem that may not be immediately visible, affecting energy and behavior first.
A wall map shows the number of insertions per hole across the enrichment wall, while charted data displays time at wall for each elephant, and which elephant was engaging with the wall when.
Handlers can view each elephant within the herd, and quickly view a more detailed overview of statistics on their engagement.
Handlers can document what the current day's enrichment wall setup is through a photo upload feature. By checking the data against the enrichment setup, handlers can better determine insights from the relationship.
The final iteration of the device was left installed for approximately eight days, before there was a connection failure between the primary microcontroller and one of the detection nodes.
To evaluate our results, we used video to monitor how often the elephants came up to the wall, and for how long they stayed engaging with it. If the elephant was standing near the wall, but not actively engaging with it, it was not counted. Video also let us see which elephant was interacting with the wall, and therefore further code the data to see individual characteristics of each individual.
Overall, the total duration of engagement with the wall increased 71% while total number of approaches increased 176%. When discussing the results of the data with the handlers, they had mentioned they placed the elephants on a diet during the period of which our installation was up, meaning they were putting less food behind the wall, the elephants' primary reason for being there in the first place. From our data we were pleased to see a significant increase in engagement with the wall across all elephants, and especially pleased knowing these elephants were described as being not interested in acoustic enrichment when tried before. This is a promising start to develop further technology-enabled enrichment for elephants and other zoo animals. We hope that by establishing baselines with data collected from enrichment products, that they can be used to detect significant changes in the animals behavior, and therefore provide early warning signs of unseen health issues.
With results of our installation complete, we compiled the project and data into a research paper, accepted into and presented at the annual Animal Computer Interaction international conference. The Animal Lab at Georgia Tech has continued to develop research relations with Zoo Atlanta following this project, designing conceptual enrichment and welfare evaluation devices.
Arianna Mastali is continuing work with the elephants at Zoo Atlanta while pursuing her PhD under Dr. Melody Jackson. Currently, she is working on a computer vision model that can detect trunk insertions with just a camera rather than proximity sensors. This will significantly reduce the infrastructure requirements while increasing the ease of scale so that the entire enrichment wall can be analyzed, rather than the four holes of fourteen we instrumented here.
Thank you to Dr. Melody Jackson and the Animal Lab at Georgia Tech, Charles Ramey, Zoo Atlanta staff and elephant handlers, and design shop managers Noah Posner and Tripp Edwards for the incredible support and guidance.
