Industries / Media, arts and recreation / Zoos and aquariums
Animal care
Daily husbandry (feeding, housing, training and handling) and the work of judging whether each animal is doing well, which zoos call animal welfare.
5
Milestones achieved, from 2011 to 2016.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2011
Keepers step behind barriers
In "free contact", keepers share space with elephants and steer them with a bullhook (a rod with a hooked tip); in "protected contact", gates, posts and fences separate keeper and elephant, and training uses rewards. Oakland Zoo introduced protected contact in 1991, after an elephant named Smokey killed trainer Lorne Jackson. On 12 August 2011 the Association of Zoos and Aquariums (AZA), the main accrediting body in the United States, ruled that "no later than September 1, 2014", keepers at its members "shall not share the same unrestricted space with elephants", except for listed tasks such as health procedures, transport and calf management.
LessonThe rule came after a long run of injuries; PETA, an advocacy group, counted 15 people killed and more than 135 injured by captive elephants in the United States over 20 years.
Credit
Oakland Zoo; AZA
2015
A welfare model for all
WAZA's Animal Welfare Strategy asks zoos and aquariums everywhere to judge welfare with the Five Domains model: four physical domains (nutrition, environment, physical health and behaviour) and one mental domain, which together give "insight into the overall welfare status". The strategy has nine chapters of recommendations and checklists.
LessonA target named the "2023 Animal Welfare Goal" was amended in November 2024: associations needed a welfare evaluation process by November 2024, and member zoos have up to five years from that process's approval to go through it.
Credit
WAZA
2016
Elephant welfare measured
Nine linked studies of African and Asian elephants in 68 accredited North American zoos, which the authors call "the largest prospective zoo-based animal welfare study conducted to date", looked at abnormal behaviour, foot and joint problems, lying down, walking and breeding health. "Social and management factors were found to be important for multiple indicators of welfare, while exhibit space was found to be less influential than expected."
LessonMore space alone was not the answer; who an elephant lived with and how it was managed mattered more.
Credit
C. L. Meehan, J. A. Mench, K. Carlstead and J. N. Hogan (PLOS ONE), funded by an Institute of Museum and Library Services grant to the Honolulu Zoological Society
2016
SeaWorld stops orca breeding
On 17 March 2016 SeaWorld Entertainment said it "will end all orca breeding as of today", so its orcas would be the last in its parks, and it began a partnership with the Humane Society of the United States. Its chief executive, Joel Manby, wrote that the company had not collected an orca from the wild in almost four decades. Fortune reported boycotts and a fall in ticket sales after the documentary Blackfish, released three years earlier; attendance had dropped 4% in 2014.
LessonManby presented the decision as a response to an "attitudinal change" among the public, a change he said SeaWorld had helped to create.
Credit
SeaWorld Entertainment (Joel Manby)
2016
A shared behaviour app
ZooMonitor, a tablet app built by Lincoln Park Zoo in Chicago for recording what animals do and where in their enclosure they do it, was released in 2016. By January 2025 more than 1,400 organizations in over 70 countries had registered, its data had appeared in more than 50 research papers, and the zoo made it free for everyone. A shared-data feature let the zoo study giraffes at 18 US zoos at once.
Credit
Lincoln Park Zoo (an idea of its chief executive, Megan R. Ross); developer Tracks Data Solutions
Watch for
AI watches animals at night
The University of Surrey and Marwell Wildlife, a zoo near Southampton in England, say they will build "the first AI health and welfare monitoring system focused on zoo animals": cameras and machine-learning software that flag unusual night-time behaviour for keepers. It will be tested first with giraffes and red river hogs during a three-year project funded with more than £344,000 through Innovate UK.
Achieved whenthe partners publish, for at least a year of use, how many of the system's alerts matched problems that keepers or vets later confirmed, and how many such problems it missed.
Claimed by
University of Surrey and Marwell Wildlife, May 2026
Real-time stress tracking
Emirates Park Zoo and Resort in Abu Dhabi is "preparing to introduce AI powered systems to track animal behaviour in real time", to help keepers read patterns in health, stress and activity.
Achieved whenthe zoo or an outside party reports which species are monitored and how the system's readings compare with keepers' own records.
Claimed by
Emirates Park Zoo, reported by Khaleej Times, December 2025
Open
01
A pacing detector for zoos
Software that reads footage from cameras a zoo already has and reports each day how long each animal spent pacing or swaying. Welfare scientists call these repeated, unvarying movements stereotypic behaviour and treat them as a warning sign. The detector would be trained and tested on clips from many zoos and species, and the labelled clips published.
The briefCredit · unclaimed02
Space-use maps without observers
A tool that turns ordinary enclosure cameras into the reports ZooMonitor produces by hand, around the clock: a heat map of where each animal spends its time, and an activity budget (the share of time spent resting, moving, feeding and so on).
The briefCredit · unclaimed
Open 01 · the brief
A pacing detector for zoos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Software that reads footage from cameras a zoo already has and reports each day how long each animal spent pacing or swaying. Welfare scientists call these repeated, unvarying movements stereotypic behaviour and treat them as a warning sign. The detector would be trained and tested on clips from many zoos and species, and the labelled clips published.
Why it is still open
Each published system covers a few animals in one place. Zürl and colleagues (Ecological Informatics, 2024) tracked two polar bears over 607 days, classified pacing with 94.9% accuracy and published their code. Lund and colleagues (Animals, 2024) coded seven nights of two African elephants and found their models "had certain limitations regarding detection of complex behaviors, such as the stereotyped behavior sway". Surrey and Marwell describe their 2026 project as the first AI welfare-monitoring system for zoo animals.
Built with what exists
CCTV footage that zoos already record; Ultralytics YOLO11 to find animals in each frame; DeepLabCut, which Lund and colleagues used, for body posture; the Zürl team's public tracking code as a starting baseline; Label Studio for labelling clips.
First step
Ask two zoos that store night-time CCTV for one week of footage from a bear or big-cat enclosure, with the keepers' notes for that week.
Done looks like
At least 500 hours of labelled footage from at least five zoos and five species (for example bears, big cats and elephants), published; on footage from zoos not used in training, the detector's daily pacing minutes fall within 10% of two trained human observers' totals on at least 80% of animal-days.
Open 02 · the brief
Space-use maps without observers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that turns ordinary enclosure cameras into the reports ZooMonitor produces by hand, around the clock: a heat map of where each animal spends its time, and an activity budget (the share of time spent resting, moving, feeding and so on).
Why it is still open
ZooMonitor's heat maps and activity budgets come from "trained data collectors" who "closely observe animals" in person. The only automated version found in this search mapped the positions of two polar bears in one enclosure.
Built with what exists
Existing CCTV; YOLO11 with a multi-object tracker such as ByteTrack; OpenCV's homography functions to map camera pixels onto an enclosure plan, the method Zürl and colleagues used; ZooMonitor exports as the human comparison.
First step
Pick one enclosure where staff already run ZooMonitor sessions, save the camera feed during ten of them, and compare the two by hand.
Done looks like
For 100 observation sessions covering at least three species at two zoos, the automated heat maps and activity budgets agree with ZooMonitor sessions recorded at the same time to within 10 percentage points for each zone and each behaviour, with the comparison published.