How does a rover know where it is on the Moon?
There is no GPS on the Moon. One way for a rover to find itself is to recognise a place it has seen before. That is hard on the lunar surface, because every crater field looks alike and the shadows move with the Sun. LunarSim is our open benchmark for this problem, and Galatea is our method for solving it.
The same location under seven Sun positions
These eight images show the same spot. The first is the reference “map” image. The other seven were taken from almost the same position, but with the Sun somewhere else. On the Moon there is no air to soften shadows, so moving the Sun changes almost everything the camera sees.
Sun position given as azimuth / elevation in degrees. Rendered with LunarSim.
Test yourself against the rover’s AI
In Where on the Moon? you see what a rover’s camera sees and pick the matching place from six map images. The levels get harder as the Sun moves further from where it was when the map was made. After each round you see what our AI chose.
The rover AI picks the right place in 81–96% of these rounds, depending on the lighting. A strong AI trained on Earth photos manages only 52–84%. How do people compare? That is an open question, and one we hope the game can help answer.
Start a missionSee how sunlight affects recognition
Drag the Sun around a lunar scene and watch the shadows move. For every Sun position, the Sun Lab shows whether our model still recognises the place among 150 map images.
Open the Sun Lab