PLA missile scientists say the accuracy of hypersonic weapons could be improved by more than 10 times if control is taken out of human hands and given to a machine. Their paper, published last week in the peer-reviewed journal Systems Engineering and Electronics, proposes using artificial intelligence to write the weapon’s software “on the fly” through a unique flight control algorithm as it travels at hypervelocity. Professor Xian Yong and Li Bangjie, from Rocket Force Engineering University’s college of war support, said more decision-making power would be handed to the smart weapon – giving its human controllers no idea how it would behave after the launch button was pressed – but overall positioning accuracy “would increase by one to two orders of magnitude”. Chinese air force may have cracked how to land a hypersonic drone Conventional missiles are equipped with positioning software which is installed and fine-tuned on the factory floor. But if the software was written by AI, with a different algorithm for each weapon, the researchers found they could address the challenges of controlling flight at five times the speed of sound or beyond. Whether a hypersonic weapon can hit its target after travelling hundreds or thousands of kilometres depends heavily on how precisely it can determine its own position while making complex manoeuvres during flight. At hypervelocity, parts of an aircraft can get hotter than the sun’s surface, breaking air molecules into electrically charged ions which form a plasma coating. This reduces the craft’s radar signature but can also make it blind and deaf – unable to pick up GPS signals or use other references, such as the Earth’s magnetic field, for guidance. These extreme conditions over long distances have forced a reliance on built-in inertial sensors – such as quartz accelerometers and laser gyroscopes – which can only estimate a hypersonic weapon’s location. This is despite sophisticated control software and painstaking on-the-ground testing. The researchers said physical disturbances to the sensors were inevitable during their assembly, transport and routine maintenance. And each time the weapon is powered up, it affects the hardware, causing further deviations from the factory settings. Xian and Li’s team believe factory settings could eventually be scrapped for good with the application of AI. It would require considerable computing power but was feasible with current technology, the researchers said. Their study showed an AI-based system could keep a hypersonic weapon on course with an accuracy of about 10 metres (32 feet). Using their method, the AI would start work immediately after launch, before the weapon reached hypervelocity, to calculate its position using the signal from the GPS or BeiDou – China’s navigational positioning system – and compare it with the results generated by the on-board sensors to evaluate the actual condition of the hardware. Based on this fresh information, the AI would create a unique positioning algorithm for the weapon’s flight control programme before it entered the cruising stage of hypersonic flight. China designs hypersonic jet bigger than Boeing 737 with wings like Concorde In one simulated flight, the AI-generated algorithm underwent thousands of rounds of evolution during the initial stage of flight on a 10-year-old Intel Xeon CPU. The final version was obtained in about 20 seconds. The speed of processors used in China’s hypersonic weapons programme remains classified, but their performance has been increasing steadily, according to the researchers. Chinese scientists have used artificial intelligence to address other aspects of hypersonic flight, including engine control and communication. While China has fielded various types of hypersonic weapons, civilian applications of the technology remain challenging. In May, Chinese space authorities announced plans to build a small passenger plane, capable of reaching anywhere on Earth in an hour, by 2035. This would require it to reach a speed of Mach 15.