When Should Robots Take Risks Exploring Other Worlds?

On May 1st, 2009, after five years on the Martian surface, the Spirit rover got stuck in a patch of soft sand (where it would remain for the rest of its mission). On February 13th, 2019, NASA officials declared that Spirit’s sister – the Opportunity rover – had concluded its mission after a planetary dust storm forced it into hibernation mode about seven months prior. And in March 2017, the Curiosity rover’s wheels showed signs of their first break, thanks to years of traveling over rough terrain. Such are the risks of sending rover missions to other planets in search of discoveries that can lead to scientific breakthroughs.

But what constitutes an acceptable risk for a robotic mission, and when are mission controllers justified in taking them? As it turns out, a pair of researchers from the Robotics Institute‘s School of Computer Science at Carnegie Mellon University (CMU) in Pittsburgh have developed a new approach for weighing the risks against the scientific value of sending planetary rovers into dangerous situations. The researchers are now working with NASA to implement their approach for future robotic missions to the Moon, Mars, and other potentially-hazardous environments in the Solar System.

The research team included David Wettergreen, a research professor with the RI, and Alberto Candela, a former robotics Ph.D. student with the RI and a current data scientist at NASA’s Jet Propulsion Laboratory. The paper that describes their approach, titled “An Approach to Science and Risk-Aware Planetary Rover Exploration,” was presented by Wettergreen and Candela at the IEEE and RSJ International Conference on Intelligent Robots and Systems – which took place from October 23rd to 27th, in Kyoto, Japan.

Robotic missions measure scientific value based on their confidence in interpreting mineral data from scanning rocks. If it concludes that it has correctly identified the mineral composition of rocks without the need for additional measurements, it may decide to explore somewhere else. If its confidence is low, it might decide to keep studying the current area to improve the accuracy of its readings. For their new approach, Wettergreen and Candela combined models that weigh the scientific value of the region against any potential hazards to the rover.

As Wettergreen, who has worked on autonomous planetary exploration for decades at Carnegie Mellon University, summarized in a CMU press release:

“We looked at how to balance the risk associated with going to challenging places against the value of what you might discover there. This is the next step in autonomous navigation and to producing more and better data to aid scientists.”

To measure risk, Wettergreen and Candela relied on a model that combines data on the topography and material makeup of the local terrain to determine how difficult it will be for the rover to reach its destination. For instance, sloped terrain with loose sand (a major concern on Mars) would present a high level of risk, as the rover might trigger a slide as it attempted to ascend the slope and end up buried. This is precisely what happened to the Spirit rover in 2004 when it became stuck in a dune, and its wheels slipped when it tried to move.

The team tested their framework using a simulation based on real Mars surface data. By navigating a simulated rover through this terrain, they charted different paths based on varying risks, then evaluated the scientific obtained by these missions. “The rover did very well on its own,” said Candela. “Even under high-risk simulations, there were still plenty of areas for the rover to explore, and we found that we still made interesting discoveries.”

This new approach builds on work dating back to the 1980s, where researchers have proposed and demonstrated methods that would allow rovers to navigate across the surface of other planets. This includes Ambler, a six-meter (~20 foot) tall, six-legged robot developed by scientists at the CMU that was tested in the 1990s. This robot demonstrated how rover missions could prioritize their goals and chart their own paths in extraterrestrial environments, which inspired additional robotic testbeds.

Examples include Ratler, a four-wheeled, skid-steered robot developed by the Sandia National Laboratories as a testbed for lunar navigation software. This was followed by Nomad, a demonstration rover tested in the Atacama Desert in the summer of 1997. Then there was Hyperion, a project led by Wettergreen that built a rover designed for Sun-Synchronous Robotic Exploration (SSRE) – where a robot follows the sun to keep its solar-powered batteries charged. Since 2004, researchers at CMU have used the Zoë rover as a testbed for autonomous navigation and exploration technologies.

This included a previous version of the method developed by Wettergreen and Candela. As of 2012, Zoë has conducted tests in the Atacama Desert, where it traveled hundreds of kilometers to test systems for autonomous exploration and sample collection. In 2013, the rover decided to drill at a site that led to the discovery of highly-specialized microbes, thus demonstrating how automatons systems can result in valuable scientific returns. In the future, Candela and Wettergreen hope to use Zoë to test their new method in the Utah desert. As Wettergreen said:

“Our goal is not to eliminate scientists, not to eliminate the person from the inquiry. Really, the point is to enable a robotic system to be more productive for scientists. Our goal is to collect more and better data for scientists to use in their investigations.” 

They also anticipate that their research could be invaluable to future lunar exploration, which includes NASA’s long-awaited return to the Moon (the Artemis Program). In anticipation of sending crewed missions to the lunar surface for the first time since the Apollo Era, robotic missions need to investigate the local terrain, scout resources, and assess potential dangers to astronauts. Scientists could use Wettergreen and Candela’snew approach as a tool to map out potential routes in advance and balance the risk of traveling them with the potential for major scientific finds.

Their approach could also assist next-generation rovers sent to distant locations where continuous human involvement is impractical. This includes astrobiology missions to Europa, Titan, and other bodies that could reveal evidence of life beyond Earth. For missions closer to home, autonomous systems that can assess risk would also free up mission controllers to focus on interpreting scientific data.

Further Reading: Carnegie Mellon University, IEEE Xplore