Drones have a wide range of uses, but sending them into unfamiliar environments can be difficult. Knowing how to navigate previously unseen (or significantly altered) environments for drones to effectively complete tasks such as delivering packages, monitoring wildlife, or conducting search and rescue missions is important. Researchers at the Massachusetts Institute of Technology (MIT) believe they have found a more effective way to help drones fly through uncharted space, thanks to liquid neural networks.
In 2021, MIT created a liquid neural network inspired by the adaptability of the organic brain. Artificial intelligence and machine learning algorithms can learn and adapt to new data in the real world, not just during training. In other words, they can think on the fly.
They can make sense of information important to the drone’s task while ignoring irrelevant features of the environment, the researchers note. It is also possible to dynamically capture the true causal relationship of the tasks that have been performed. science robotics. This is “the key to robust performance of liquid networks under distribution shifts.”
Liquid neural nets outperformed other approaches to navigation tasks, the researchers said in their paper. The algorithm “showed excellent ability to make reliable decisions in uncharted territory such as forests, cityscapes, and environments with added noise, rotation, and occlusion,” the university said in a press release.
MIT points out that deep learning systems can fail to understand causality and cannot always adapt to different environments and conditions. This poses a problem for drones that must be able to react quickly to obstacles.
“Our experiments show that we can effectively train drones to locate objects in the forest in the summer and deploy the model in the winter. , and can also be used in urban environments with various tasks such as seeking and following.” Daniela Rus, Director of the Institute for Science and Artificial Intelligence (CSAIL), Professor and co-author of the paper at MIT In a statement, he said, “This adaptability is made possible by the causal foundation of our solution. These flexible algorithms may support decision-driven decision making in the future. It covers time-varying data streams such as medical diagnostics and self-driving applications.”
Researchers trained the system on data captured by human pilots. This has allowed us to explain the ability of pilots to use their navigational skills in new environments with vastly changed conditions and landscapes. I discovered that I could track it. They suggest that limited data from expert sources combined with an improved ability to understand new environments could make drone operations more reliable and efficient.
Dr. Alessio Lomuscio, Professor of AI Safety, said: (Computing Department) of Imperial College London. “In this context, the performance reported in this study of liquid neural networks, a new brain-inspired paradigm developed by the authors at MIT, is noteworthy.”
