While generative models have seen significant adoption across a wide range of data modalities, including 3D data, a consensus on which model is best suited for which task has yet to be reached. Further, conditional information such as text and images to steer the generation process are frequently employed, whereas others, like partial 3D data, have not been thoroughly evaluated. In this work, we compare two of the most promising generative modelsDenoising Diffusion Probabilistic Models and Autoregressive Causal Transformers-which we adapt for the tasks of generative shape modeling and completion. We conduct a thorough quantitative evaluation and comparison of both tasks, including a baseline discriminative model and an extensive ablation study. Our results show that (1) the diffusion model with continuous latents outperforms both the discriminative model and the autoregressive approach and delivers state-of-the-art performance on multi-modal shape completion from a single, noisy depth image under realistic conditions and (2) when compared on the same discrete latent space, the autoregressive model can match or exceed diffusion performance on these tasks.

Shape completion, i.e., predicting the complete geometry of an object from a partial observation, is highly relevant for several downstream tasks, most notably robotic manipulation. When basing planning or prediction of real grasps on object shape reconstruction, an indication of severe geometric uncertainty is indispensable. In particular, there can be an irreducible uncertainty in extended regions about the presence of entire object parts when given ambiguous object views. To treat this important case, we propose two novel methods for predicting such uncertain regions as straightforward extensions of any method for predicting local spatial occupancy, one through postprocessing occupancy scores, the other through direct prediction of an uncertainty indicator. We compare these methods together with two known approaches to probabilistic shape completion. Moreover, we generate a dataset, derived from ShapeNet [1], of realistically rendered depth images of object views with ground-truth annotations for the uncertain regions. We train on this dataset and test each method in shape completion and prediction of uncertain regions for known and novel object instances and on synthetic and real data. While direct uncertainty prediction is by far the most accurate in the segmentation of uncertain regions, both novel methods outperform the two baselines in shape completion and uncertain region prediction, and avoiding the predicted uncertain regions increases the quality of grasps for all tested methods. Web: https://github.com/DLR-RM/shape-completion

In this work, a new ensemble method for the task of category recognition in different environments is presented. The focus is on service robotic perception in an open environment, where the robot’s task is to recognize previously unseen objects of predefined categories, based on training on a public dataset. We propose an ensemble learning approach to be able to flexibly combine complementary sources of information (different state-of-the-art descriptors computed on color and depth images), based on a Markov Random Field (MRF). By exploiting its specific characteristics, the MRF ensemble method can also be executed as a Dynamic Classifier Selection (DCS) system. In the experiments, the committee- and topology-dependent performance boost of our ensemble is shown. Despite reduced computational costs and using less information, our strategy performs on the same level as common ensemble approaches. Finally, the impact of large differences between datasets is analyzed.

This paper presents a method to transfer functional grasps among objects of the same category through contact warping and local replanning. The method transfers implicit knowledge that enables an action on a class of objects for which no explicit grasp or task information has been given in advance. Contact points on the source object are warped based on global and local shape similarities to the target object. These warped contacts are then used to define a hand posture that reaches close to them, while at the same time provides the desired functionality on the object. The approach is tested on different sets of objects with a success rate of 87.5%, and large benefits are shown when compared to a naive technique that only transfers a suitable hand pose to the novel object.

This video presents the recent upgrades of the mobile humanoid Agile Justin, bringing it closer to an ideal platform for research in autonomous manipulation. Significant upgrades have been made in the fields of mechatronics, 3D sensors, tactile skin, massive GPGPU based computing power, and software communication framework. In addition, first algorithms and two experimental scenarios are presented that take advantage of these new capabilities.

Object detection is here considered as the problem of retrieving from scene data segments that belong to objects from the sought category. The method proposed and investigated works with dense range data, as can be acquired with low-cost sensors. It does not require any training, but just a single geometric prototype that may be taken from an internet repository. Experiments with various household and office scenes are reported, and the performance is quantified on a public dataset. One of the tested variants achieves an F-score and average precision of 94% at total recall, and a correct nearest-neighbor rate of 97%.

Multi-fingered robot grasping is a challenging problem that is difficult to tackle using hand-coded programs. In this paper we present an imitation learning approach for learning and generalizing grasping skills based on human demonstrations. To this end, we split the task of synthesizing a grasping motion into three parts: (1) learning efficient grasp representations from human demonstrations, (2) warping contact points onto new objects, and (3) optimizing and executing the reach-and-grasp movements. We learn low-dimensional latent grasp spaces for different grasp types, which form the basis for a novel extension to dynamic motor primitives. These latent-space dynamic motor primitives are used to synthesize entire reach-and-grasp movements. We evaluated our method on a real humanoid robot. The results of the experiment demonstrate the robustness and versatility of our approach.

We present a method for transferring grasps between objects of the same functional category. This transfer is intended to preserve the functionality of a grasp constructed for one of the objects, thus enabling the analogous action to be performed on a novel object for which no grasp has been specified. Manipulation knowledge is hence generalized from a single example to a class of objects with a significant amount of shape variability. The transfer is achieved through warping the surface geometry of the source object onto the target object, and along with it the contact points of a grasp. The warped contacts are locally replanned, if necessary, to ensure grasp stability, and a suitable grasp pose is computed. We present extensive results of experiments with a database of four-finger grasps, designed to systematically cover variations on grasping the mugs of the Princeton Shape Benchmark.

A method is presented for non-rigid alignment of a source shape to a target shape through estimating and interpolating point-wise correspondences between their surfaces given as point clouds. The resulting mapping can be non-smooth and non-isometric, relate shapes across large variations, and find partial matches. It does not require a parametric model or a prior of deformations. Results are shown for some objects from the Princeton Shape Benchmark and a range scan.

This video presents a humanoid two-arm system developed as a research platform for studying dexterous two-handed manipulation. The system is based on the modular DLR-Lightweight-Robot-III and the DLR-Hand-II. Two arms and hands are combined with a three degrees-of-freedom movable torso and a visual system to form a complete humanoid upper body. The diversity of the system is demonstrated by showing the mechanical design, several control concepts, the application of rapid prototyping and hardware-in-the-loop (HIL) development as well as two-handed manipulation experiments and the integration of path planning capabilities.

The thalamus is the major gate to the cortex, and its contribution to cortical receptive field properties is well established. Cortical feedback to the thalamus is, in turn, the anatomically dominant input to relay cells, yet its influence on thalamic processing has been difficult to interpret. For an understanding of complex sensory processing, detailed concepts of the corticothalamic interplay need to be established. To study corticogeniculate processing in a model, we draw on various physiological and anatomical data concerning the intrinsic dynamics of geniculate relay neurons, the cortical influence on relay modes, lagged and nonlagged neurons, and the structure of visual cortical receptive fields. In extensive computer simulations, we elaborate the novel hypothesis that the visual cortex controls via feedback the temporal response properties of geniculate relay cells in a way that alters the tuning of cortical cells for speed.

We present a method for object recognition and pose estimation from noisy range data as provided by stereo processing. From the range dates, points of high surface curvature are estimated. By comparing three-point geometric relations, hypothetical correspondences are established between data and model points of high curvature. The hypothetical correspondences give rise to pose hypotheses which are evaluated with respect to the raw range data using a crude surface model. We show examples that demonstrate the method's tolerance to noise and occlusion.