Using Pokémon GO Data for Delivery Robots
Almost ten years after Pokémon GO transformed our world into an augmented reality playground, the vast data gathered from millions of players is now being repurposed to improve how autonomous delivery robots navigate urban streets.
Niantic Spatial, the team behind the popular game, has joined forces with Coco Robotics, a company that focuses on short-distance delivery robots for groceries and food. This collaboration will utilize Niantic’s Visual Positioning System, a navigation technology built from over 30 billion images collected by the game. This system enables delivery robots to maneuver through sidewalks and urban environments with remarkable accuracy.
The Visual Positioning System can reportedly locate objects within a few centimeters by analyzing nearby buildings and landmarks. This represents a considerable step forward compared to traditional GPS methods. The crowdsourced mapping initiative is one of the largest data collection projects ever conducted through a mobile gaming app, showcasing the potential of user-generated content long after it was first gathered.
John Hanke, CEO of Niantic Spatial, mentioned in a conversation with MIT Technology Review that making Pikachu realistically navigate the world and having Koko’s robot operate safely in that same space are essentially similar challenges.
When Pokémon GO launched in 2016, it quickly became a cultural sensation, peaking at around 230 million monthly active users. The game compelled players to physically visit various locations and use their phone cameras to search for virtual creatures in real-life settings. While its popularity has declined, estimates suggest it still has about 50 million active players.
In 2020, Niantic took its data collection to the next level by introducing a feature called Field Research, which encouraged players to scan real-world landmarks in exchange for in-game rewards. This process generated 3D models of real locations, capturing various conditions, from different angles to differing weather scenarios.
Unlike GPS, which relies on satellite signals, visual positioning systems base location on the surrounding environment. This method addresses some major flaws found in traditional GPS, especially in urban areas where tall buildings can disrupt satellite signals and affect accuracy.
Hanke remarked that while the potential for last-mile delivery robots is substantial, navigating busy city streets is still a significant engineering challenge.
Coco’s delivery robot employs a visual positioning system along with four cameras to accurately perceive its surroundings. This technology aims to ensure that robots deliver food and groceries promptly, attempting to solve issues that have plagued other autonomous systems, particularly on college campuses where robots sometimes get lost or face difficulties crossing streets.
This reuse of Pokémon GO data highlights a growing trend, where information originally collected for one reason is repurposed years later for entirely different applications. Notably, Google’s CAPTCHA test, which asks users to identify images to prove they are human, has been suspected of training AI models. Similarly, law enforcement has reportedly accessed or purchased user-generated content from mapping tools like Waze to assist with investigations.
