MIT researchers suggest their findings could lead to more reliable and efficient drone operations.
Drones have a wide range of applications, but sending them into unfamiliar environments can be a challenge. Whether delivering a package, monitoring wildlife or conducting search and rescue missions, knowing how to navigate previously unseen surroundings (or ones that have changed significantly) is critical for a drone to effectively complete tasks. Researchers at the Massachusetts Institute of Technology (MIT) believe they’ve found a more effective way of helping 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. The artificial intelligence and machine learning algorithms are able to learn and adapt to new data in the real world, not only while they’re being trained. They can think on the fly, in other words.
They’re able to understand information that’s critical to a drone’s task while dismissing irrelevant features of an environment, the researchers note. The liquid neural nets can also “dynamically capture the true cause-and-effect of their given task,” according to a paper published in Science Robotics. This is “the key to liquid networks’ robust performance under distribution shifts.”
The liquid neural nets outperformed other approaches to navigation tasks, the researchers noted in the paper. The algorithms “showed prowess in making reliable decisions in unknown domains like forests, urban landscapes and environments with added noise, rotation and occlusion,” the university said in a press release.
MIT points out that deep learning systems can flounder when it comes to understanding causality and can’t always adapt to different environments or conditions. That poses a problem for drones, which have to be able to react quickly to obstacles.
“Our experiments demonstrate that we can effectively teach a drone to locate an object in a forest during summer, and then deploy the model in winter, with vastly different surroundings, or even in urban settings with varied tasks such as seeking and following,” Computer Science and Artificial Intelligence Laboratory (CSAIL) director, MIT professor and paper co-author Daniela Rus said in a statement. “This adaptability is made possible by the causal underpinnings of our solutions. These flexible algorithms could one day aid in decision-making based on data streams that change over time, such as medical diagnosis and autonomous driving applications.”
The researchers trained their system on data captured by a human pilot. This enabled them to account for the pilot’s ability to use their navigation skills in new environments that have undergone significant changes in conditions and scenery. In testing the liquid neural nets, the researchers found that drones were able to track moving targets, for instance. They suggest that marrying limited data from expert sources with an improved ability to understand new environments could make drone operations more reliable and efficient.
“Robust learning and performance in out-of-distribution tasks and scenarios are some of the key problems that machine learning and autonomous robotic systems have to conquer to make further inroads in society critical applications,” says Alessio Lomuscio, PhD, professor of AI Safety (in the Department of Computing) at Imperial College London. “In this context the performance of liquid neural networks, a novel brain-inspired paradigm developed by the authors at MIT, reported in this study is remarkable.”