Autonomous drones could transform agriculture – but without GPS, indoor environments remain a major barrier. A La Trobe University collaboration at Innovation Central Melbourne is tackling that challenge, developing a greenhouse-ready drone system built for real-world farming.
When drones move indoors, they lose their sense of place. Inside greenhouses and controlled-environment farms, GPS signals don’t reach. Plants are often densely packed, access is limited, and manual inspection is slow and inconsistent. For growers who rely on frequent, high-quality data to monitor crop health, that gap matters. Missed stress, disease, or nutrient issues can quickly translate into lost yield.
This was the challenge behind Drones Indoors, a project led by La Trobe University’s Institute of Sustainable Agriculture and Food (LISAF) in collaboration with Innovation Central Melbourne and Polybee, a Singapore-based agri-tech company. The goal was practical rather than theoretical – to build a safe, accurate drone system that could operate autonomously inside a greenhouse and deliver data that could be used in day-to-day operations.
Anthony D’Agata, Technical and Marketing Engineer of Plant Phenotyping at La Trobe University, was assigned as project coordinator through the Department of Education’s Regional Research Collaboration Program.
“One aspect of the program was to aid in the proof-of-concept for automated flight and image capture using a DJI Mavic 3M drone in a greenhouse environment,” Mr D’Agata says.
At the centre of the challenge was localisation – how to tell a drone exactly where it is, to within a few centimetres, in a complex indoor environment.
From markers to meaningful data
The project adapted Polybee’s proprietary computer-vision technology to automate the flight of off-the-shelf drones inside GPS-denied environments. By recognising known features within the greenhouse, the system builds a map of the space and tracks the drone’s position in real time, allowing it to follow pre-planned flight paths safely and accurately. The focus of the work was to tailor this technology to LISAF’s greenhouses and assess how well the drones could capture high-resolution images for plant phenotyping.
In parallel, La Trobe researchers are also building their own indoor drone capability through Innovation Central Melbourne. “The team has developed an autonomous indoor drone mission stack that mirrors industry-leading architectures, integrating localisation, control and computer-vision capabilities,” says Jeff Jones, Director of Innovation Central Melbourne. “The system is designed to support autonomous waypoint missions for agricultural and other monitoring applications, strengthening La Trobe’s in-house capacity alongside its industry collaborations. ICM mange the process including detail design and best-practice IP management.”
In practice, that capability matters because access is limited inside greenhouses. “High plant density makes it difficult for humans to access and survey crops effectively,” Mr D’Agata explains. “A drone provides rapid results in hard-to-reach places compared to manual inspections, enabling frequent, detailed assessments of plant health, fruit detection, and nutrient status.”
Localisation was paired with a navigation and control stack that defined safe flight paths above crops, enforced height and speed limits, and enabled emergency stops. A mission-planning dashboard allowed researchers to visualise flights in real time through a digital twin of the greenhouse, while the drone also captured images that could later be analysed to understand plant health and other desirable traits relevant to agriculture.
Getting the system to fly reliably indoors was not straightforward. “The biggest challenges were ensuring reliable localisation within two to five centimetres accuracy, maintaining safe flight constraints above plants, and dealing with environmental factors like airflow turbulence from HVAC systems,” says Mr D’Agata. “We worked through these by increasing the number of markers, refining speed and height limits, testing emergency stops, and adjusting camera angles to reduce plant disturbance.”
From prototype to deployment
For Polybee, the collaboration delivered more than technical validation. “The La Trobe team brought genuine curiosity and rigour to a challenge that’s central to what Polybee does,” says Sarthak Mittal, Lead Product Engineer of Robotics at Polybee. “They didn’t just deliver research; they worked alongside us to turn it into something we could actually deploy in a glasshouse.”
That focus on deployment mattered. Matt Ru, an ICM software engineer on the project, describes it as a lesson in building systems that survive contact with reality.
“I loved the challenge of building the entire system from the ground up and making sure it survived the transition from a whiteboard idea to a live business tool,” Mr Ru says. “The most valuable lesson was learning how to be creative when facing hardware constraints, ensuring the final automation pipeline wasn’t just a lab experiment, but a commercially viable solution.”
By the end of the project, the team had demonstrated safe, repeatable autonomous flight indoors, with localisation accuracy suitable for precision phenotyping. Just as importantly, the system was designed to scale. Once a facility is mapped and analytics pipelines are established, multiple drones can be deployed across greenhouses of almost any size.
For La Trobe, the project highlights the value of university–industry collaboration when it is grounded in real constraints and real outcomes.
“This was a genuine partnership,” Mr D’Agata says. “It brought together plant science, engineering, and industry expertise to create a platform that can support future research and real-world agricultural deployment.”
