Skip to content Skip to navigation

SLAB Generates SPEED+ Dataset for Training Autonomous Optical Navigation Algorithms

By Adam Koenig   July 27, 2021


SLAB PhD student Jeff Park has begun generating datasets that will be used to train next-generation autonomous optical navigation algorithms and enable new rendezvous and proximity operations capabilities. The data will be made publicly available to the astrodynamics and space engineering community in the form of SPEED+: a new spacecraft pose estimation dataset for machine learning applications. Pose labels are accurate to millimeters in position and milli-degrees in attitude and will form the basis for a new upgraded international competition organized with the European Space Agency.

https://damicos.people.stanford.edu/sites/g/files/sbiybj2226/f/speedplus_clip_1.mp4

The SPEED+ dataset is generated using two Kuka robotic arms, a model of a target spacecraft, and custom-build illumination devices.

Adam Koenig is a Postdoctoral Scholar at SLAB