We propose a real-time RGB-based pipeline for object detection and 6D pose estimation. Our novel 3D orientation estimation is based on a variant of the Denoising Autoencoder that is trained on simulated views of a 3D model using Domain Randomization. This so-called Augmented Autoencoder has several advantages over existing methods: It does not require real, pose-annotated training data, generalizes to various test sensors and inherently handles object and view symmetries. Instead of learning an explicit mapping from input images to object poses, it provides an implicit representation of object orientations defined by samples in a latent space. Experiments on the T-LESS and LineMOD datasets show that our method outperforms similar modelbased approaches and competes with state-of-the art approaches that require real pose-annotated images.

We introduce a scalable approach for object pose estimation trained on simulated RGB views of multiple 3D models together. We learn an encoding of object views that does not only describe an implicit orientation of all objects seen during training, but can also relate views of untrained objects. Our single-encoder-multi-decoder network is trained using a technique we denote "multi-path learning": While the encoder is shared by all objects, each decoder only reconstructs views of a single object. Consequently, views of different instances do not have to be separated in the latent space and can share common features. The resulting encoder generalizes well from synthetic to real data and across various instances, categories, model types and datasets. We systematically investigate the learned encodings, their generalization, and iterative refinement strategies on the ModelNet40 and T-LESS dataset. Despite training jointly on multiple objects, our 6D Object Detection pipeline achieves state-of-the-art results on T-LESS at much lower runtimes than competing approaches. 1

We propose a real-time RGB-based pipeline for object detection and 6D pose estimation. Our novel 3D orientation estimation is based on a variant of the Denoising Autoencoder that is trained on simulated views of a 3D model using Domain Randomization.This so-called Augmented Autoencoder has several advantages over existing methods: It does not require real, pose-annotated training data, generalizes to various test sensors and inherently handles object and view symmetries. Instead of learning an explicit mapping from input images to object poses, it provides an implicit representation of object orientations defined by samples in a latent space. Our pipeline achieves stateof-the-art performance on the T-LESS dataset both in the RGB and RGB-D domain. We also evaluate on the LineMOD dataset where we can compete with other synthetically trained approaches.We further increase performance by correcting 3D orientation estimates to account for perspective errors when the object deviates from the image center and show extended results. Our code is available here 1

Teams of mobile robots will play a crucial role in future missions to explore the surfaces of extraterrestrial bodies. Setting up infrastructure and taking scientific samples are expensive tasks when operating in distant, challenging, and unknown environments. In contrast to current single-robot space missions, future heterogeneous robotic teams will increase efficiency via enhanced autonomy and parallelization, improve robustness via functional redundancy, as well as benefit from complementary capabilities of the individual robots. In this article, we present our heterogeneous robotic team, consisting of flying and driving robots that we plan to deploy on scientific sampling demonstration missions at a Moon-analogue site on Mt. Etna, Sicily, Italy in 2021 as part of the ARCHES project. We describe the robots' individual capabilities and their roles in two mission scenarios. We then present components and experiments on important tasks therein: automated task planning, high-level mission control, spectral rock analysis, radio-based localization, collaborative multi-robot 6D SLAM in Moon-analogue and Marslike scenarios, and demonstrations of autonomous sample return.