The development and use of robotics, drone technology and AI have mostly taken place in the Global North. However, it is the Global South that may stand to benefit most from these new tools. WeRobotics is a global initiative of more than 20 innovation labs across Africa, Latin America, Asia and Oceania called the Flying Labs network, set up to explore this opportunity. Their mission is to decrease the digital divide and use the Flying Labs to build skills, expertise and an innovation ecosystem to address local challenges. They primarily use drone solutions to acquire high‑resolution local aerial data to support the implementation and impact of local aid, health, development and environmental projects. However, analysing the high‑dimensional drone footage is labour- and time-intensive, and most existing AI tools have not been developed to detect the features most needed by the Flying Labs.

In recent years, WeRobotics has run three challenge prize competitions called the Open AI Challenge, which have led to the development of AI tools to help analyse drone footage in Tanzania, the Caribbean and the South Pacific. Understanding local context can be vital to building an AI tool that is able to detect the most relevant features on maps, so the team takes additional steps to encourage participation from local innovators, as well as the international data science and AI community. For the Open AI Tanzania Challenge in 2018, they partnered with Black in AI, DataKind and local universities to attract African data scientists. The winning entry was developed by tuning existing convolutional neural networks (deep learning) to perform segmentation and classification of building types appropriate for the local context.

The Open AI Tanzania Challenge attracted 126 participants and the AI classifier is being integrated into the global WeRobotics analysis platform so that other Flying Labs in the network can choose to use it for analysing their own images. In 2019, the Open AI Caribbean Challenge – which focused on developing AI to accurately identify rooftop material from drone footage – had 2,753 entries from 1,425 participants. The winning entry significantly improved on the existing benchmarks and will help local humanitarian and disaster relief workers to anticipate which buildings are most vulnerable to destruction.