FocusSPACE ROBOTS

Robotic space exploration agents

+ See all authors and affiliations

Science Robotics  21 Jun 2017:
Vol. 2, Issue 7, eaan4831
DOI: 10.1126/scirobotics.aan4831

Abstract

By making their own exploration decisions, robotic spacecraft can conduct traditional science investigations more efficiently and even achieve otherwise impossible observations, such as responding to a short-lived plume at a comet millions of miles from Earth.

To explore the vast reaches of space, we send robots to places where the environment is too hostile for humans (e.g., Jovian radiation) and/or it is far more cost-effective to use robots (e.g., Earth- or Mars-orbiting spacecraft). Frequently, these space robots are out of communication with Earth and their human operators due to the extreme distances and light-time communication delays (e.g., 11-hour round trip to the New Horizons spacecraft at the Kuiper belt). Further, the high cost and over-subscription of the ground communication stations that must service all missions throughout the solar system limit the number of contacts each spacecraft receives. For example, Earth orbiters typically get six communication passes per day, whereas Mars rovers get only one.

One goal of autonomy is to enable robots to detect and respond to unexpected conditions without sitting idle until the next Earth command arrives. Remote science operations with Mars rovers are pre-planned and constrained to the local area visible from the previous day’s downlink. In an exciting development, many spacecraft have increasing ability to make their own decisions and accelerate scientific discoveries.

For robots to conduct autonomous science investigations, they must first be able to detect and characterize features of interest. In some cases, these features can be defined ahead of time. For example, Earth orbiters analyze images as they are collected to detect unusual but important events, such as volcanic activity (1), fires, or floods. They use machine-learning classifiers to distinguish between snow, water, and ice to track seasonal changes in Arctic regions (2). Change detection has been used to detect and track dust devils on the surface of Mars (3).

Increasing autonomy and onboard decision-making for orbiters and rovers enables them to also detect new unforeseen features. Earth orbiters use statistical measures of salience (or “surprise”) to detect unusual features (4), and the same methods are being tested for use in remote, poorly understood environments, such as the surface of Europa. Similar technologies have been demonstrated on plume data from comet Tempel 1 and Enceladus, a moon of Saturn (5). The Mars Science Laboratory rover (Curiosity) regularly selects new, interesting rock targets and autonomously fires its ChemCam laser to collect compositional spectra (6). Of particular interest are rare dynamic features—such as dust devils on the surface of Mars or active jets erupting from primitive bodies such as the comet 67P/Churyumov-Gerasimenko. Robotic spacecraft can detect these events and take action by collecting new images and data, enabling new kinds of science even across the vastness of space and in unfamiliar environments.

Onboard analysis can also interpret data content to make the most of limited onboard storage and data downlink capacity. Onboard analysis of thermal signatures in Hyperion data enabled the Earth Observing One orbiter to send only kilobytes of thermal data instead of 1 gigabyte of raw data (1). The Mars Exploration Rover Opportunity can detect and track dust devils in real time, which enables data reduction by cropping non–dust devil portions of images (3).

An onboard scheduling system that can rewrite the observation plan based on detected events enables an even higher level of robotic intelligence. In 1999, the Remote Agent Experiment flying on the Deep Space One mission scheduled engineering operations for two periods totaling 48 hours. In 2017, the Autonomous Sciencecraft (7) completed over a dozen years of nearly continuous operations of the Earth Observing One spacecraft using both onboard and ground-based artificial intelligence (AI) scheduling, operating the mission through over 60,000 images and 2.9 million commands. The Intelligent Payload EXperiment Earth-orbiting CubeSat used onboard planning for over 1 year to achieve autonomous payload operations, including automatically conducting tens of thousands of image analyses (5). The Mars 2020 rover mission is currently developing an onboard scheduler to enable the rover to collect more science data when resources are available (8). All of these planning systems use some type of timeline modeling of the spacecraft state and resources and apply AI-based search to find satisfactory schedules.

Future exploration plans place even higher demands on autonomy. Scientists believe that our solar system contains at least eight “ocean worlds” (www.nasa.gov/specials/ocean-worlds). These liquid-bearing worlds may contain hydrothermal vents that could harbor extraterrestrial life, by analogy with extremophiles living near hydrothermal vents in the depths of Earth’s oceans. The most promising of these oceans lie beneath icy shells kilometers thick on Europa, Enceladus, and Pluto. After penetrating through these ice layers, a life-hunting submarine would need to explore for days, weeks, and even months without human direction. Adaptive data clustering could enable a future mission to downlink a map of terrain types along with exemplars instead of thousands of individual images (9).

A single autonomous robot can accomplish a lot, but groups of collaborating spacecraft have even more potential. Multiple Earth-observing spacecraft have been linked with ground sensors in sensor webs to track volcanism (10), flooding, wildfires, and other science events. In the future, orbiters, rovers, and aerial vehicles could autonomously organize and coordinate to better explore distant worlds.

The ultimate challenge for robotic science explorers would be to visit our nearest neighboring solar system, Alpha Centauri (e.g., Breakthrough Starshot). To traverse a distance of over 4 light years, an explorer to this system would likely endure a cruise of over 60 years. Upon arrival, the spacecraft would need to operate independently for years, even decades, exploring multiple planets in the system. Today’s AI innovations are paving the way to make this kind of autonomy a reality.

The Cassini spacecraft observed active plumes erupting from the surface of Enceladus, a moon of Saturn.

Although Cassini did not have the ability to allow it to autonomously respond to observation of these plumes, future mission concepts include onboard autonomy methods to detect and respond to such plumes from moons, comets, or other bodies.

Image credit: NASA/JPL-Caltech

REFERENCES AND NOTES

Acknowledgments: This work was performed by the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA.
View Abstract

Navigate This Article