Starting in May 2016, NASA’s Curiosity rover began choosing some of its own science targets on Mars using software called AEGIS - Autonomous Exploration for Gathering Increased Science - developed at NASA’s Jet Propulsion Laboratory. AEGIS lets the rover point its ChemCam instrument, a laser that vaporizes rock and reads the resulting spark with a spectrometer, at targets the rover selects on its own rather than waiting for human operators on Earth.
The motivation is the communication gap. Commands and data travel between Earth and Mars on a delay, and the rover often parks at a new spot after a drive with no time in the day’s plan for scientists to review images and pick targets before the next move. AEGIS analyzes images from the rover’s cameras, identifies geological features such as outcrops, ranks them by criteria the science team specifies, and immediately aims ChemCam at the best one - capturing measurements that would otherwise be lost. By mid-2017 it had directed ChemCam dozens of times and helped flag rocks unusually rich in chlorine and silica.
The work was reported in a 2017 paper in Science Robotics and was carried forward to the Perseverance rover’s SuperCam instrument. It is one of the clearest operational examples of onboard AI doing real science autonomously, hundreds of millions of kilometers from any human.