Pose Bowl: Pose Estimation Track

Develop algorithms for use on inspector spacecraft that take and process photos of other ships. In the Pose Estimation Track, solutions will identify the position and orientation of the spacecraft's camera across sequences of images. #science

$28,000 in prizes
may 2024
414 joined

About the problem

Space is an inherently hazardous environment, and vehicles are regularly damaged in space. Traditional methods of repairing spacecraft damage are not ideal or feasible for all situations. For example, dispatching people to make repairs can be dangerous and using robotic arms to make repairs is expensive. The danger and expense of in-space repairs are greater when the damage is not well-characterized in advance.

This comeptition is part of a broader effort to provide safe, low-cost methods of assessing and potentially repairing damaged ships in space using an inspector spacecraft (e.g., R5 spacecraft). An ideal version of an inspector spacecraft would be like a fire extinguisher: a small, cheap, standard safety device.

Inspector spacecraft - also called "chasers" - must be able to take pictures at different orientations and quickly process them in order to identify a target spacecraft and intelligently conduct an inspection of its exterior. In addition, inspector spacecraft must accomplish these tasks with very limited computing resources, and must deploy algorithms that function well for any generic target spacecraft, since the geometry of a target vehicle is not always known in advance of an inspection, or may have been changed by damage.

The goal of this challenge is to identify methods that chaser spacecraft can use to detect a target vehicle and determine changes in its own camera's pose as it moves around the target. Critically, the object detection and pose estimation solutions developed for this challenge are intended for use on R5 CubeSats that will run software on off-the-shelf computer hardware. Software for inspector spacecraft must be accurate and fast, but use very limited computing resources. Thus, the core task of this challenge is creatively applying methodologies for object detection and pose estimation problems (such as those linked below) to function well, quickly, and with limited computing resources.

Additional resources

In this challenge, you can use pretrained models and borrow from existing solutions to related problems. Below, you will find ideas and approaches that may be helpful to borrow from. Note that many of the resources below tackle similar but not identical problems, so you may need to adapt the ideas to work for this task, which involves image chains and computational constraints.

Determining the pose of an unknown object

3D Reconstruction and SLAM

These are useful ideas that can be applied to solving this problem.

NeRF and other scene rendering approaches

These specific approaches are likely too computationally expensive for this competition, but may be worth considering.

Spacecraft pose estimation with known vehicle

While different from the objective of this competition, these may be worth reading for context.