Drones have a wide range of applications, but sending them into unfamiliar environments can be challenging. Whether delivering a package, monitoring wildlife, or conducting search and rescue missions, knowing how to navigate environments never seen before (or ones that have changed significantly) is critical for a drone to complete tasks efficiently. effective. Researchers at the Massachusetts Institute of Technology (MIT) believe they have found a more effective way to help drones fly through unknown spaces, thanks to liquid neural networks.
MIT created its Liquid Neural Networks, which are inspired by the adaptability of organic brains, in 2021. AI and machine learning algorithms can learn and adapt to new data in the real world, not just while they’re being trained. They can think on the fly, in other words.
They are able to comprehend information that is critical to a drone’s task while ruling out irrelevant features of an environment, the researchers say. Liquid neural networks can also “dynamically capture the true cause and effect of their assigned task,” according to a paper published in Robotic Science. This is “the key to strong performance of liquid networks in distribution shifts.”
Liquid neural networks outperformed other approaches to navigation tasks, the researchers noted in the paper. The algorithms “showed prowess in making reliable decisions in unknown domains such as forests, cityscapes, and environments with added noise, rotation, and occlusion,” the university said in a press release.
MIT notes that deep learning systems can fail when it comes to understanding causation and cannot always adapt to different environments or conditions. That poses a problem for drones, which need to be able to react quickly to obstacles.
“Our experiments show that we can teach a drone to locate an object in a forest during the summer and then deploy the model in winter, with a very different environment, or even in urban environments with varied tasks like search and follow,” Computer said. . Director of the Science and Artificial Intelligence Laboratory (CSAIL), MIT professor and co-author of the paper, Daniela Rus, said in a statement: “This adaptability is made possible by the causal underpinnings of our solutions. These flexible algorithms could one day help in decision making based on data streams that change over time, such as medical diagnostics and autonomous driving applications.
The researchers trained their system on data captured by a human pilot. This allowed them to account for the pilot’s ability to use his navigation skills in new environments that have undergone significant changes in conditions and scenery. By testing liquid neural networks, the researchers found that drones could track moving targets, for example. They suggest that marrying limited data from expert sources with an enhanced ability to understand new environments could make drone operations more reliable and efficient.
“Robust learning and performance in out-of-the-box tasks and scenarios are some of the key issues that machine learning and autonomous robotic systems have to overcome to further advance critical applications in society,” says Alessio Lomuscio, PhD, AI Safety Professor. (in the Department of Computing) at Imperial College London. “In this context, the performance of liquid neural networks, a new brain-inspired paradigm developed by the MIT authors, reported in this study, is remarkable.”