How a $90 million robot got fooled by the Moon

An image from NASA’s Lunar Reconnaissance Observatory shows the crash site of the Hakuto-R lander.

When it all comes down to it, the robot was surprised by a crater.

When the Hakuto-R Moon lander, built by the Japanese firm ispace, was descending toward the lunar surface at 120 kmh (75 mph) on April 26, it should have been about 15 minutes away from a gentle touchdown. Instead, it would soon smash into the surface, adding another crater to the pockmarked lunar regolith.

Read more

The first-order explanation from ispace after the crash was that the vehicle ran out of fuel for the engines that slowed the vehicle. But why didn’t the vehicle have enough fuel? The answer to that lies within the software used to pilot the vehicle and a change in destination late in the mission planning process.

Without a pilot onboard, a robotic Moon lander has to pilot itself. To do so, it relies on two key sources of information.

A sense of where you are

The first source of information is an estimate of its location spit out by something called an inertial measurement unit, or IMU. These devices use gryoscopes and accelerometers to measure changes in the position of the spacecraft compared to an initial starting point throughout its flight. The second source is a laser rangefinder, which measures the actual distance between the robot and the lunar surface.

In the post-flight analysis, ispace engineers realized there was a conflict between these two sources of information: The estimated altitude over the Moon and the distance reported by the laser rangefinder differed by 5 km (3.1 miles). The robot’s guidance software was instructed to ignore the the rangefinder in case of a discrepancy, since engineers figured that was more likely to be the result of a malfunctioning sensor. Doing so, the lander burned its propellant too quickly—then spent two minutes in free fall.

A last minute change in destination

The question remains—why didn’t the software expect this disparity? According to ispace chief technology officer Ryo Ujiie, the issue came from a decision to move the robot’s target landing site from a flat plain on the Moon called Lacus Somniorum to the Atlas crater. When the lander flew over the lip of the crater, the sudden change in altitude below, measured by the laser rangefinder, was interpreted as a mistake.

A rendering of the lander’s flight path over the Moon.

Ujiie told reporters on Friday (May 26) that the decision to land in the crater was made months before the Dec. 2022 launch, and after the lander’s critical design review—essentially the final step in designing the vehicle’s systems—was complete. Simulations designed to test to the lander’s software simply did not take into account the specific topography of the landing site, leading to the conflicted measurements and the failure.

Asked if the vehicle might have landed at the original site, Ujiie said that “it is very hypothetical, [but] yes, we might have had a chance to successfully land on the Moon.” The change in destination was prompted by the desire to maximize the benefit of the mission for the onboard payloads; the crater was considered more scientifically interesting than the plains, which can be more easily examined by telescope.

Try, try, again

Ispace lost 100 million Japanese yen ($711,000) in expected customer payments due to the failure of mission, but CEO Takeshi Hakamada said the company expects to receive a payment from its insurer on the mission, Mitsui Sumitomo, though he could not say the amount or the cost of the policy.

The company is gearing up for two more attempts to reach the Moon, one next year with a lander based on the Hakuto-R design, and another in 2025 in collaboration with Draper Labs, a US company with a contract to deliver scientific instruments to the Moon for NASA. Draper designed the landing software used for this mission, but ispace executives said the problems the lander encountered stemmed from the requirements and parameters controlled by their mission management team.

For the next mission, Ujiie says that the company will fine-tune the lander software with more specific terrain simulations and incorporate better sensor data. The company will also determine the final landing target and simulation requirements further in advance.

“We believe that this is our commitment and our duty to all our stakeholders— ‘Never Quit the Lunar Quest!’” Hakamada said in a statement.

More from Quartz

Sign up for Quartz's Newsletter. For the latest news, Facebook, Twitter and Instagram.

Click here to read the full article.