56-gram drone gets ant-inspired AI eyes to navigate autonomously

Researchers have developed an autonomous navigation system for small, light drones inspired by insects.

A team at TU Delft was motivated by biological discoveries on how ants use their ability to see their surroundings and calculate their steps to navigate safely back home.

According to engineers, robots may go vast distances using this method and return home with minimal computation and memory (0.65 kiloByte per 100 m).

“In the future, tiny autonomous robots could find a wide range of uses, from monitoring stock in warehouses to finding gas leaks in industrial sites,” said researchers in a statement.

Insect-inspired “snapshot” navigation

Tiny robots , weighing from ten to a few hundred grams, hold significant potential for real-world applications. Their lightweight design ensures safety, even in accidental collisions, and their small size allows them to navigate narrow areas. If produced affordably, they can be deployed in large numbers, efficiently covering vast areas like greenhouses for early pest or disease detection.

However, autonomous operation is challenging due to limited resources compared to larger drones. Navigation is particularly problematic. While GPS can aid outdoor navigation, it is ineffective indoors and inaccurate in cluttered environments. Indoor wireless beacons are costly and impractical in scenarios like search-and-rescue.

According to researchers, most AI for autonomous navigation is designed for large robots, using heavy, power-intensive sensors like LiDAR, which are unsuitable for tiny robots. Vision-based approaches, though power-efficient, require creating detailed 3D maps, demanding substantial processing power and memory beyond the capacity of small robots.

Researchers turned to nature and took inspiration from insects for tiny robot navigation, using minimal resources. Insects combine odometry (tracking motion) with visually guided behaviors (view memory).

In the “snapshot” model, insects like ants periodically capture snapshots of their surroundings. When near a snapshot, they compare current visuals, minimizing differences to navigate precisely back to the snapshot and correct odometry drift.

Drone’s efficient indoor navigation

The DU Helft team modified previously developed techniques to develop a bio-inspired approach. This strategy combines visual homing, which directs orientation in relation to visual cues in the environment, with odometry, which measures the distance traveled along a specific direction.

The researchers tested their method in several indoor conditions using a 56-gram Crazyflie Brushless drone with a panoramic camera, microcontroller, and 192 kB of memory.

Initially, the robot took off and flew toward its target, stopping periodically to take pictures of its surroundings. The drone employed visual homing to travel the same path back, regularly making course corrections for drift by comparing its current location with waypoint photos.

The approach was incredibly memory-efficient because of the pictures’ high compression and precise spacing. All visual processing happened on a tiny computer called a “micro-controller”, which can be found in many cheap electronic devices.

According to the team, the proposed strategy is less versatile than state-of-the-art methods, lacking mapping capability but enabling a return to the starting point, which is adequate for many applications.

Drones might fly out, collect data, and then return to the base station for applications like crop monitoring in greenhouses and warehouse stock tracking. Images pertinent to the objective could be saved on a little SD card and processed later by a server. However, they wouldn’t require them to use simple navigation.

The details of the team’s study were published in the journal Science Robotics .