Today, we'll share more about how AI also is used to help human operators drive VIPER and create highly accurate maps of the rover's mission area on the Moon.

Like a self-driving car, VIPER has cameras that monitor the environment around the rover and software that detects hazardous locations where it shouldn't go. However, unlike self-driving cars, this software isn't on board the rover; it's back on Earth, and presents its conclusions to the rover drivers who use this information, along with many other sources, to decide how the rover should move.

One reason AI isn't completely given the reins to the VIPER mission, is that AI techniques require a lot of training data - and this is the first time NASA will be remotely driving a robotic rover on the Moon. Using AI while always keeping humans in the loop provides a balance of risk and reward by using innovative and efficient techniques while avoiding unnecessary risk.

"VIPER is using AI as a tool; we're not giving it the keys to the car," said Mark Shirley, who created the original deterministic planner for VIPER at NASA's Ames Research Center in California's Silicon Valley. "And for this science mission, we don't have to - the Moon is close enough that we can monitor these systems that are still learning this new environment and watch everything, like how you'd want to watch over a new driver."

We don't know everything about the environment of the Moon, but we do know a lot - and we can use AI to help us fill in the blanks.

Learning the Terrain
Planning routes and sensing hazards aren't the only ways VIPER is using artificial intelligence. Other AI techniques are helping generate very high-resolution terrain maps. Most of our data about the Moon comes from LRO (Lunar Reconnaissance Orbiter), including several hundred photographs of VIPER's mission area and topographical data obtained by shooting a laser down at the lunar surface and seeing how long it took to bounce back up.

A subfield of AI, called computer vision, can determine what the local slope is at each pixel using points of altitude, images, and our knowledge of the lunar environment, including how lunar regolith reflects light, where the Sun is in relation to the Moon, what direction the camera is facing, and how bright each pixel is.

All those slopes can be combined to create a terrain model that helps the VIPER team know the shape of the lunar surface. This shape can be used to calculate how the shadows move as the sun moves, and these moving shadows inform SHERPA's - short for the System Health Enabled Real-time Planning Advisor - route planning. It is especially important to know how the shadows move because VIPER runs on solar power. Being stuck in a shadow for too long could be deadly for the rover.

All these pieces fit together. The high-resolution terrain maps created from LRO data generate maps of moving shadows, which SHERPA accounts for planning VIPER's route. Temporal constraint techniques help mesh activities on the ground with activities on board the rover. Finally, the hazards pointed out automatically from the rover's camera images help the VIPER team navigate the minute-to-minute decisions that come up while exploring another world.

As AI continues to develop as a field, many of its methods will end up becoming part of the regular toolkit for engineers and scientists. VIPER uses some of the current well-trodden techniques, while also pushing the boundaries of AI's applications. In the case of SHERPA, the cutting-edge techniques come from a subfield of AI called decision making under uncertainty. This will be the first time these techniques are used on a space mission, and if successful, could open the door to similar AI approaches being deployed on other missions to worlds beyond our own.

Related Links
VIPER
Mars News and Information at MarsDaily.com
Lunar Dreams and more