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
Completed may 2024
414 joined

Why

Space is an inherently hazardous environment, and vehicles are regularly damaged in space. Conducting in-space repairs can be dangerous and expensive, whether by dispatching people or by using robotic arms. Inspecting and characterizing the damage in advance is critical to mitigating the danger and expense of performing repairs.

This competition 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. An ideal version of an inspector spacecraft would be like a fire extinguisher: a small, cheap, standard safety device.

The Solution

In the Pose Bowl Challenge, solvers helped advance space inspection technology by competing to develop solutions for in-space inspection that would work on any type of target spacecraft. Solutions were developed against severe constraints on compute resources to reflect the limitations of the compute hardware available on a spacecraft. The data for this challenge were simulated images of spacecraft taken from a nearby location in space, as if from the perspective of a chaser. In the Object Detection track, solvers developed solutions to detect target spacecraft depicted in an image. In the Pose Estimation track, solvers worked to determine the relative pose of the chaser camera across images of the target spacecraft taken as the chaser moves around it.

The Results

Over 800 participants from around the world competed in this difficult, highly technical challenge. Across the challenge tracks, solvers submitted 1,625 solutions. In the Object Detection track, solutions easily outperformed the benchmark model. Top solvers narrowed in on YOLOv8 as the best approach for low-resource object detection. The Pose Estimation track was a true challenge—there was limited progress on improving model accuracy during the competition, but solvers were able to explore interesting directions in modeling approaches that will inform future research in pose estimation.

See the results announcement for more on the prize-winning approaches and the teams behind them. The top solutions from this competition are linked below and available for anyone to use and learn from.


RESULTS ANNOUNCEMENT + MEET THE WINNERS

WINNING SUBMISSIONS ON GITHUB