How do insects manage to go so far from their home and still return? The answer to this question is relevant not only for biology but also for the development of artificial intelligence for small, autonomous robots. Researchers at TU Delft were inspired by biological discoveries about how ants visually recognize their environment and combine it with step counting to safely return home. They used these insights to create an insect-inspired autonomous navigation strategy for small, lightweight robots. This strategy enables robots to return home after long journeys with extremely low computational resource and memory consumption (1.16 kilobytes per 100 meters). In the future, small autonomous robots could be used for a wide range of tasks, from monitoring inventory in warehouses to detecting gas leaks in industrial plants. The researchers published their findings in the journal Science Robotics on July 17, 2024.

Support for small robots
Small robots, ranging from a few tens to a few hundred grams, have potential for many interesting real-world applications. Due to their light weight, they are extremely safe even if they accidentally hit someone. Because of their size, they can move in tight spaces. If they can be produced cheaply, they can be deployed in larger numbers, quickly covering a large area, for example, in greenhouses for early detection of pests or diseases. However, enabling such small robots to operate autonomously is difficult because, compared to larger robots, they have extremely limited resources.

One of the main obstacles to using small robots is their ability to navigate autonomously. Robots can get help from external infrastructure, such as GPS outdoors or wireless communication transmitters indoors. However, relying on such infrastructure is often undesirable. GPS is not available indoors and can be very inaccurate in urban canyons. Installing and maintaining transmitters indoors is expensive or simply impossible, for example, in search and rescue scenarios.

The artificial intelligence needed for autonomous navigation with limited resources has been developed with large robots in mind, such as autonomous cars. Some approaches use heavy, energy-intensive sensors such as LiDAR laser distance meters, which small robots cannot carry or power. Other approaches use visual sensors, which are very energy-efficient and provide rich information about the environment. However, these approaches usually try to create very detailed 3D maps of the environment. This requires large amounts of processing and memory, which only large and energy-intensive computer systems can provide, too large for small robots.

Step counting and visual cues
That's why some researchers have turned to nature for inspiration. Insects are particularly interesting because they operate at distances that could be relevant for many real-world applications, while using very scarce perception and computational resources. Biologists are increasingly understanding the underlying strategies that insects use. Specifically, insects combine tracking their own movement (so-called "odometry") with visually guided behavior based on their low-resolution but almost omnidirectional visual system (so-called "visual memory"). While odometry is increasingly understood down to the neural level, the precise mechanisms behind visual memory are still less known. Therefore, there are different theories about how insects use vision for navigation. One of the earliest theories proposes a "snapshot" model. According to this model, an insect like an ant occasionally takes a snapshot of its surroundings. Later, when it approaches the shooting location, the insect can compare its current visual impression with the snapshot and move to minimize the differences. This allows the insect to navigate, or 'home', to the shooting location, eliminating any drift that inevitably occurs when relying solely on odometry.

"Snapshot-based navigation can be compared to how Hansel tried not to get lost in the fairy tale of Hansel and Gretel. When Hansel was throwing stones on the ground, he could find his way home. However, when he was throwing breadcrumbs that birds ate, Hansel and Gretel got lost. In our case, the stones are snapshots," says Tom van Dijk, the first author of the study, "Like stones, for a snapshot to work, the robot must be close enough to the shooting location. If the visual environment differs too much from the shooting location, the robot can go in the wrong direction and never return. Therefore, it is necessary to use enough snapshots – or in the case of Hansel, throw enough stones. On the other hand, throwing stones too close together would quickly deplete Hansel's stones. In the case of the robot, using too many snapshots leads to high memory consumption. Previous work in this area usually had snapshots very close together, so the robot first visually 'homes' to one snapshot, then to the next."

"The main insight of our strategy is that you can space the snapshots much further apart if the robot travels between snapshots based on odometry," says Guido de Croon, professor at TU Delft and co-author of the paper, "Navigation will work as long as the robot ends up close enough to the snapshot shooting location, i.e., as long as the robot's odometry drift falls within the snapshot area. This also allows the robot to travel much further, as the robot flies much slower when moving towards a snapshot than when flying from one snapshot to the next based on odometry."

The proposed insect-inspired navigation strategy enabled the 56-gram "CrazyFlie" drone, equipped with an omnidirectional camera, to cover distances of up to 100 meters with a consumption of only 1.16 kilobytes. All visual processing took place on a small computer called a "microcontroller," which can be found in many inexpensive electronic devices.

Application of robotic technology
"The proposed insect-inspired navigation strategy is an important step towards applying small autonomous robots in the real world," says Guido de Croon, "The functionality of the proposed strategy is limited compared to the most advanced navigation methods. It does not generate a map and only allows the robot to return to the starting point. Nevertheless, for many applications, this may be more than enough. For example, for inventory monitoring in warehouses or crop monitoring in greenhouses, drones could fly, collect data, and then return to the base station. They could store images relevant to the mission on a small SD card for post-processing on a server. However, these images would not be needed for navigation itself."

Additionally, in the future, this technology could also be used for other purposes such as infrastructure monitoring or aiding in rescuing the injured. For example, in natural disaster scenarios, small robots could quickly search through rubble and find survivors, significantly increasing the chances of timely rescue. Also, in industrial plants, these robots could regularly monitor pipelines and equipment, identifying potential problems before they develop into serious failures. The combination of minimal energy consumption and high efficiency makes this technology extremely promising for a wide range of applications.

Source: Delft University of Technology