New dual-resolution technique opens door for faster drone exploration
A team of researchers from Carnegie Mellon University has successfully developed a new dual-mapping technique that could help robots explore areas faster and more efficiently. By producing both a site's high- and low-resolution map, this new technique enables robots to explore areas using only a fraction of the computing power typically needed for a similar task.
More efficient exploration
In fact, according to the study published in Science Robotics, during simulations and real-world experiments with single robots, this method was 80% more efficient and had a 50% lower computational run time than existing strategies. But why is this important?
Imagine autonomous robots, like self-driving cars or drones, as tourists in a new city. To explore efficiently, they need good maps. Currently, these robots generate very detailed (high-resolution) maps, which take up a lot of computational power and time. It's similar to attempting to navigate a city using an excessively detailed street map that displays every building, tree, and sidewalk. This approach is not very effective since the robots only require this level of information for their immediate surroundings, not for the entire city.
Or, if you prefer gaming analogies, the new technique is akin to sandbox games, where you only need high-resolution graphics for where the player is currently in the world. Anything "over the horizon," so to speak, can be very low resolution or not even rendered.
Chao Cao and his team have devised a fresh approach to address the issue. They have programmed the robots to generate two types of maps: a detailed map of the immediate vicinity and a simpler map of a wider area. This is similar to how a tourist would use a detailed map to navigate their current location and a simpler map to plan their next destination.
Using a two-map strategy, robots can explore their surroundings with greater thoroughness and efficiency. Tests have shown that this new method is more efficient and has a significantly lower computational run time than existing methods. This approach works effectively for both single robots and teams of robots, as well as for both ground and aerial robots.
Won DARPA award
This strategy won the “Most Sectors Explored Award” during the Defense Advanced Research Projects Agency (DARPA) Subterranean Challenge, a top competition for the autonomous exploration of tunnels, caves, and industrial sites. The robots could also share information efficiently, even in environments that are typically difficult for communication, like caves.
You can read the study for yourself in the journal Science Robotics.
Study abstract:
"By creating both low-resolution and high-resolution maps, a new visualization strategy enables autonomous robots to explore their surroundings thoroughly. During simulations and real-world experiments with single robots, this method was 80% more efficient and had a 50% lower computational run time than existing strategies. These improvements were also evident with multi-robot teams and ground and aerial robots. When used by a team of three ground robots, the approach won the DARPA Subterranean Challenge – a flagship competition for autonomous exploration of tunnel systems, caves, and industrial sites. Autonomous robots often rely on vision sensors to map surroundings and scope out targets. Yet, current visualization strategies are not incredibly efficient.
Because they typically resort to generating high-resolution maps, the techniques consume computational power and time. Here, Chao Cao and colleagues designed a two-pronged visualization framework designed to optimize how single robots and robot teams accomplish missions. One step generates a high-resolution map for detailed analyses of robots’ immediate surroundings, which is useful for gathering fine-scale details of the local environment. The other step builds a lower-resolution map that encompasses a broader space, which can help guide the robots to explore additional sites of potential interest. Furthermore, Cao et al.’s framework also supports the efficient transfer of information among robot teams in cave environments and other restricted spaces that can pose communication challenges. Their system helped Autonomous robots achieve the 'Most Sectors Explored Award' during the DARPA Subterranean Challenge."