Autonomous AI Navigation in NASA’s OSIRIS-REx and DART Missions

NASA’s OSIRIS-REx and DART missions revolutionized deep-space exploration by utilizing autonomous optical navigation powered by artificial intelligence to reach their asteroid targets. Because vast distances create significant communication delays between Earth and deep space, human controllers could not pilot these spacecraft in real-time during their final approaches. Instead, both missions relied on advanced, onboard machine vision and autonomous navigation algorithms to process live camera feeds, identify hazards or targets, and fire thrusters independently. This article explores how OSIRIS-REx used its Natural Feature Tracking system to safely collect a sample from asteroid Bennu, and how DART utilized its SMART Nav system to intentionally collide with the asteroid moonlet Dimorphos.

The Need for Autonomous Optical Navigation When spacecraft operate millions of miles from Earth, radio signals take several minutes to travel one way. For OSIRIS-REx, the signal delay was over 18 minutes; for DART, it was roughly 38 seconds. At these distances, relying on human operators to steer a spacecraft during high-stakes, final-minute maneuvers is impossible. Traditional navigation relies on radio tracking from Earth, which gets a spacecraft close to its target, but final terminal guidance requires optical navigation—using onboard cameras to visually lock onto the target and make immediate trajectory adjustments using artificial intelligence.

OSIRIS-REx and Natural Feature Tracking (NFT) When NASA’s OSIRIS-REx spacecraft arrived at asteroid Bennu, mission planners discovered the surface was not a smooth plain as expected, but a hazardous landscape covered in massive, jagged boulders. The safe landing zone—a crater named Nightingale—was only 52 feet wide, roughly the size of a few parking spaces, and surrounded by building-sized rocks.

To safely touch down and collect a sample without crashing, engineers developed an autonomous AI system called Natural Feature Tracking (NFT). Prior to the descent, the spacecraft extensively mapped Bennu’s surface, storing a catalog of high-resolution 3D images and terrain data in its onboard memory.

During the final four-hour descent in October 2020, the NFT system took complete control:

Visual Comparison: As OSIRIS-REx descended, onboard cameras took real-time images of Bennu’s surface. The AI compared these live images to its stored onboard map, identifying specific boulders and craters as landmarks.
Trajectory Updates: By tracking the relative position of these natural features, the spacecraft calculated its exact position, velocity, and orientation in real time.
Hazard Avoidance: The AI continuously predicted the exact point of touchdown. Had the system detected that the spacecraft was drifting toward a hazardous boulder, it was programmed to autonomously fire its thrusters and wave off the descent entirely.

Thanks to NFT, OSIRIS-REx navigated past the hazards, touched down within three feet of the target center, and successfully collected a pristine asteroid sample.

DART and SMART Nav While OSIRIS-REx used autonomy for a gentle touch-and-go maneuver, the Double Asteroid Redirection Test (DART) used AI to achieve a high-speed, precision impact. In September 2022, DART’s mission was to slam into Dimorphos, a 530-foot-wide moonlet orbiting a larger asteroid named Didymos, at 14,000 miles per hour to test planetary defense capabilities.

The major challenge for DART was that Dimorphos was too small and faint to be seen by the spacecraft's single instrument, the DRACO camera, until the final hour of the mission. During the last four hours before impact, DART handed control over to its onboard AI system: Small-body Maneuvering Autonomous Real-Time Navigation (SMART Nav).

SMART Nav executed the impact through a rapid, autonomous process:

Target Differentiation: Using advanced image-processing algorithms, SMART Nav analyzed the pixel data from the DRACO camera to spot the faint dot of Dimorphos and separate it from the much brighter glare of the parent asteroid, Didymos.
Centroid Calculation: Once locked onto Dimorphos, the AI calculated the visual center of mass (the centroid) of the irregularly shaped moonlet.
Terminal Guidance: Over the final 50 minutes, SMART Nav issued constant commands to the spacecraft's thrusters, steering DART away from the larger asteroid and directly into the center of Dimorphos.

Because DART was moving at four miles per second, the AI had to process images, calculate trajectories, and execute thruster burns within fractions of a second. It stopped adjusting its course just two minutes before impact, successfully striking within 55 feet of the asteroid’s center and measurably altering its orbit.

The Future of Deep Space AI The successes of OSIRIS-REx and DART proved that autonomous optical navigation is no longer experimental, but essential for modern space exploration. By enabling spacecraft to see, think, and react to dynamic environments without human intervention, NASA has laid the groundwork for future autonomous landings on the Moon, Mars, and the icy moons of the outer solar system.