This post is also available in: עברית (Hebrew)
Researchers in China made UAVs communicate via “group chats” to discuss and assign work to one another, much like human teams.
The research was done by a team from the School of Artificial Intelligence, Optics and Electronics at Northwestern Polytechnical University in China, led by Li Xuelong. The university stated in a “WeChat” post that the study optimally utilized large language models like ChatGPT “to life” by integrating them into actual applications.
The researchers claim that each drone was equipped with a “human brain” that allowed them to communicate with one another using natural language, an ability which according to the study was created using a Chinese open-source big language model called InternLM. “The drones showcased key abilities, including humanlike dialogue interaction, proactive environmental awareness, and autonomous entity control,” said the report.
As reported by the South China Morning Post, the researchers published a video example of how a team of five drones successfully identified a set of keys in an outdoor park. In the video, after a user gave the drones their mission, three immediately “volunteered” their search talents while two others outfitted with grippers notified the group they could recover the keys. The drone cluster decided on the job division completely independently, and after the keys were located, the drones shared images with the user to confirm their achievement.
According to Interesting Engineering, the drones have different sensors and advanced algorithms that are designed for low-altitude exploration, real-time obstacle avoidance, and precise visual positioning. The design lets the drones comprehensively observe their environment from various perspectives and positions, thus facilitating efficient data collection and task execution. These capabilities are often described as “proactive environmental awareness” and enable the drones to not only grasp their surroundings but also adjust to them as necessary.
The key searching task was divided among the four drones with each assigned a particular area to investigate, and during the operation the drones collaborated to ensure efficient coverage of their designated zones. They created a simplified terrain map as a navigational aid and exhibited the capability to detect and evade human operators in their flight path, therefore enhancing safety.
This technology appears extremely versatile, and has potential applications in security inspections, disaster relief operations, and the optimization of drone-based transport and logistics services.