You probably have seen a drone swarm televised during a New Year’s celebration. A drone swarm occurs when several drones fly in sync, and they frequently fly outside because it is challenging for the flying robots to navigate in tighter environments without hitting each other. Now, it seems that researchers at Caltech may have a way for these flying robots to fly indoors.
Caltech engineers have designed a new machine-learning algorithm to control the movement of several flying robots through cluttered, unmapped spaces, so these drones do not run into environmental obstacles or each other. The new machine-learning algorithm is called GLAS (Global-to-Local Safe Autonomy Synthesis), and the system works by providing each flying robot a degree of independence that allows it to adapt to a changing environment.
Typical swarm navigation models work by relying on existing maps of a specific area or the routes of other drones in the swarm. However, Global-to-Local Safe Autonomy Synthesis has each robot learning how to fly a given space on its own even as it coordinates with other machines. The new decentralized system helps the robots to improvise. It also makes scaling the drone swarm easier, as the computing is spread across several flying robots.
While the main machine learning allows the drones to navigate their environment, and an additional tracking controller known as Neural-Swarm helps the flying robots compensate for aerodynamic interactions, such as the downward force of air coming from machine flying overhead. The new system is more sophisticated as compared to many controllers available as they do not account for aerodynamics.
Drone swarms are mostly used for entertainment, however, they are just too helpful and too versatile to be only used for drone light shows. Drone swarms could actually help with more vital operations. There is another field which will certainly see improvements with the new model- Search and rescue. For instance, using the new software, first responders could deploy a drone swarm to quickly as well as efficiently cover an area.
Moreover, the new program could be helpful in areas that do not include flying a robot. For example, we could integrate the new tech into self-driving cars which will help us to avoid traffic jams and collisions. If the technology is included in every car, cars will be able to interact with each other and adjust accordingly to avoid traffic jams or crashes. It may take several years before programs like GLAS are implemented into real-world swarms.
Read next: This new photo-manipulation AI can transform images like a pro
Caltech engineers have designed a new machine-learning algorithm to control the movement of several flying robots through cluttered, unmapped spaces, so these drones do not run into environmental obstacles or each other. The new machine-learning algorithm is called GLAS (Global-to-Local Safe Autonomy Synthesis), and the system works by providing each flying robot a degree of independence that allows it to adapt to a changing environment.
Typical swarm navigation models work by relying on existing maps of a specific area or the routes of other drones in the swarm. However, Global-to-Local Safe Autonomy Synthesis has each robot learning how to fly a given space on its own even as it coordinates with other machines. The new decentralized system helps the robots to improvise. It also makes scaling the drone swarm easier, as the computing is spread across several flying robots.
While the main machine learning allows the drones to navigate their environment, and an additional tracking controller known as Neural-Swarm helps the flying robots compensate for aerodynamic interactions, such as the downward force of air coming from machine flying overhead. The new system is more sophisticated as compared to many controllers available as they do not account for aerodynamics.
Drone swarms are mostly used for entertainment, however, they are just too helpful and too versatile to be only used for drone light shows. Drone swarms could actually help with more vital operations. There is another field which will certainly see improvements with the new model- Search and rescue. For instance, using the new software, first responders could deploy a drone swarm to quickly as well as efficiently cover an area.
Moreover, the new program could be helpful in areas that do not include flying a robot. For example, we could integrate the new tech into self-driving cars which will help us to avoid traffic jams and collisions. If the technology is included in every car, cars will be able to interact with each other and adjust accordingly to avoid traffic jams or crashes. It may take several years before programs like GLAS are implemented into real-world swarms.
Read next: This new photo-manipulation AI can transform images like a pro