BackgroundAnimal pollinators provide vital ecosystem services that sustain biodiversity and support agricultural yields. Declining pollinator populations threaten to affect human health outcomes by decreasing crop yields and human nutrient intake, and there is a well-recognised need to understand how environmental stressors such as pesticides affect bees, the most important animal pollinators. Neonicotinoid pesticides in particular, the most widely used class of insecticide globally, pose a major threat to bees. While there is now strong evidence that neonicotinoids can affect bee behaviour and colony growth, major knowledge gaps remain in understanding how these pesticides affect pollinator health. First, the specific mechanisms by which neonicotinoid exposure affect bee colonies are not well understood, limiting the ability to predict their effects. Second, it is unclear how pesticide exposure interacts with other environmental stressors, which could either amplify or attenuate the effects of neonicotinoids. MethodsWe describe the development of an automated, robotic system aimed at helping to fill these gaps in knowledge. Specifically, this system harnesses advances in computer vision to track the behaviour of hundreds of individual bees across dozens of colonies simultaneously. Specifically, this system uses a computer-controlled, motorised camera array to collect video data from up to a dozen bumblebee colonies in parallel. We then use computer vision-based techniques to automatically measure key aspects of behaviour (eg, activity level, task performance, and social interactions) for individually tagged bees before and after exposure to neonicotinoid pesticides. FindingsUsing a preliminary version of this system, we have found that exposure to field-realistic levels of a common neonicotinoid pesticide (imidacloprid) affects key aspects of social behaviour in bumblebees (Bombus impatiens). These effects include reduced rates of nursing (27% reduction; p<0·001) and interactions with nestmates (–10·6 unique interactions per h; p<0·001), as well as a reduction in activity level within the nest (27% reduction; p<0·001), suggesting a previously unknown mechanism by which pesticide exposure might lead to impaired growth in bee colonies. InterpretationThis automated, parallel system is ideal for studying the effects of multiple environmental stressors, and we are currently extending it to investigate simultaneous exposure to neonicotinoids, thermal stress, and nutritional deficiency on the behaviour and health of bee colonies. This system can provide efficient screening of agrochemicals at low cost, and could provide an important instrument for policymakers and other shareholders to assess the risk of agrochemicals to the health of both bee and human populations. FundingWinslow Foundation, Rockefeller Foundation.