AEGIS (Autonomous Exploration for Gathering Increased Science)

Background

AEGIS (Autonomous Exploration for Gathering Increased Science) is an intelligent instrument-targeting system that allows a Mars rover to detect, prioritize, select, and observe scientific targets on its own, without waiting for input from operators on Earth. AEGIS acquires an image with a rover camera, analyzes the image to identify geological targets, chooses the targets that best match parameters specified by mission scientists, and then points a remote-sensing instrument at the selected targets, all without Earth in the loop.

AEGIS was first demonstrated on the MER rover Opportunity in 2010, where it selected targets for the Pancam imager. AEGIS was named NASA's 2011 Software of the Year for this work. Since May 2016 it has been in routine use on the MSL Curiosity rover, selecting targets for the ChemCam laser-induced breakdown spectrometer (LIBS). In 2022, AEGIS was deployed on the Mars 2020 Perseverance rover, where it autonomously targets the SuperCam remote-sensing instrument suite. AEGIS has become a routine tool that the science teams use to extend their geochemical surveys into periods when Earth cannot be in the loop.

Problem

Robotic exploration of Mars proceeds through a cyclical “ground-in-the-loop” planning process: imagery is downlinked to Earth, scientists select observation targets, and commands are uplinked to the rover on the next planning sol. Long round-trip light times, line-of-sight constraints, and limited communication windows mean that hours of usable lighting and rover power often pass at a new location before any post-drive imagery has even reached the science team.

This loss is most acute for narrow-field-of-view remote-sensing instruments such as ChemCam and SuperCam, which require precise pointing at small (centimeter-scale) features. Two compounding effects make these post-drive periods difficult to use without onboard autonomy:

1. 1. Without targeting imagery on the ground, the team cannot select specific rocks or veins for investigation, so the post-drive window either goes unused for targeted spectroscopy or is filled with non-targeted “blind” measurements that frequently miss the preferred materials.


2. 2. Mast pointing is subject to motor backlash, thermal effects, rover settling, and stereo limits, typically ~2 mrad of uncertainty. For LIBS rasters with point spacing of 1–2 mrad, this can mean missing small features such as mineral veins entirely.

An onboard system capable of recognizing geological features, ranking them against science-team priorities, and refining pointing on the order of a few milliradians can recover these otherwise lost observations, and must do so under strict constraints on flight processor time, memory, sun-safety, and rover-collision avoidance.

Technology

1. 1. Autonomous Target Selection: AEGIS analyzes a NavCam stereo source image, identifies candidate rock targets, extracts visual and geometric features (size, shape, intensity, orientation, range), filters out unsafe or unsuitable candidates, and ranks the remainder against an operator-tunable “target profile.” A visual target tracking (VTT) coordinate frame is placed on each selected target so that ChemCam or SuperCam can be pointed at it for follow-up spectroscopy.


2. 2. Pointing Refinement (“pointing insurance”): A second mode that re-analyzes a ground-targeted RMI image to correct mast pointing errors of a few milliradians, allowing very small features such as mineral veins to be hit reliably on the first attempt.


3. 3. Rockster (Rock Segmentation Through Edge Regrouping): The computer-vision component that detects rock contours in NavCam and RMI imagery using gradient-based edge detection, edge regrouping, and gap-filling. Rockster reduces millions of pixels to a small number of scientifically meaningful objects that AEGIS can rank.


4. 4. Onboard Safety Layer: AEGIS performs sun-safety filtering using operator-defined keep-in zones and the live sun ephemeris, rover-collision checking against the actual articulation state of the suspension and arm, and stereo-range feasibility checks before any pointing is committed. To date, no AEGIS-commanded action on Perseverance has triggered the rover’s flight-software safety protections.


5. 5. Resource-Aware Flight Implementation: AEGIS runs on the rover RAD750 flight processor (133 MHz, 16 MB RAM). Pointing-refinement runs complete in roughly 95 s; full NavCam targeting runs take 150–450 s, varying nearly linearly with the number of targets found. The MSL deployment added approximately 21,000 lines of code to the 3.8-million-line Curiosity flight software.


6. 6. Mars 2020 Enhancements: Retrieval of previously acquired source imagery (no need for a fresh source image), persistent “nonvolatile” AEGIS target frames retained across rover sleep cycles, deferred stereo computation saving ~120 s per run, optional Hazcam source imagery for arm-workspace targets, and ~1 mrad closed-loop pointing using vision and resolver feedback.

Impact

AEGIS has become a routine part of how NASA explores Mars and has substantially increased the rate at which both Curiosity and Perseverance produce targeted geochemical measurements:

1. - 596 MSL Curiosity targets selected as of May 11, 2026. First-year preferred-target rate >93% vs. ~24% expected from blind targeting (Francis 2017). Across the first four years of MSL operations (sols 1343–2830), AEGIS maintained >86% preferred-target selection vs. <20% blind, and outcrop/rock target rates>90% vs. <40% blind (Verma 2020).


2. - 25% increase in ChemCam observation rate. Average LIBS rate rose from 255 to 311 shots/sol after AEGIS rollout (Verma 2023). The MSL project leadership cited AEGIS as a key capability driving increased mission productivity in their extended-mission funding submission.


3. - 293 Mars 2020 Perseverance targets selected as of May 11, 2026. In an 83-sol Bunsen Peak to Bright Angel campaign, AEGIS was used in 45 of 48 uplink plans and selected at least half of all LIBS targets in 35 of those plans.


4. - Strategic decision support. AEGIS-driven SuperCam observations have informed whether the Perseverance team stays at a site or drives on. In one case AEGIS-acquired SuperCam data showed the local rocks lacked the targeted geochemistry, allowing the team to move on without expending a sampling cycle (Verma 2023).


5. - Notable discoveries. The highest chlorine concentration ever measured by ChemCam (sol 1612) and unusual phosphorus, manganese, and iron enrichment (sol 1662) were both detected on AEGIS-selected targets, prompting follow-up campaigns by the science team (Francis 2017). For SuperCam, the sol 910 AEGIS target was the first detection of silica-enriched materials formed in an ancient hydrothermal system at Jezero crater (Beck et al. 2025).

Status

• - 2010 - MER Opportunity: First operational deployment, autonomous Pancam targeting.


• - October 2015 / May 2016 - MSL Curiosity: Uplinked October 2015, in routine science use since May 2016 (sol 1343). Ongoing.


• - May 2022 / February 2023 - Mars 2020 Perseverance: Initial SuperCam/AEGIS use sol 442 (May 18, 2022); expanded deployment with adjustable parameters first used on sol 698 (February 2, 2023). Ongoing.

Development continues, focused on additional operator flexibility (source-image pointing, target type, raster shape) and on transferring autonomous-targeting concepts to future missions where round-trip light times or operational tempo make ground-in-the-loop targeting impractical.

Applications

• - MER Opportunity (2010–2018): Target selection for the panoramic camera (Pancam).


• - MSL Curiosity (since 2016): Target selection and pointing refinement for the ChemCam LIBS spectrometer and Remote Micro-Imager.


• - Mars 2020 Perseverance (since 2022): Target selection for the SuperCam LIBS and multimode remote-sensing suite.

Publications

Team

AEGIS is the product of a collaboration across multiple groups within the Artificial Intelligence & Data Science Section at the Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA.

Raymond Francis, Tara Estlin, Daniel Gaines, Gary Doran, Steve Schaffer, Vandi Verma, Michael Burl, Ellen Thiel, Benjamin Bornstein, Lauren DeFlores, Diana Blaney, Patrick Romano, Cel Skeggs, Rebecca Castaño, Susan Johnstone, Sara Montaño

Collaborators: Roger C. Wiens and Tony Nelson (Los Alamos National Laboratory); Sylvestre Maurice and Olivier Gasnault (IRAP / CNES); Jens Frydenvang (University of Copenhagen); the wider ChemCam and SuperCam science and operations teams.