Robot manipulation of rope-like objects is an interesting problem that has some critical applications, such as autonomous robotic suturing. Solving for and controlling rope is difficult due to the complexity of rope physics and the challenge of building fast and accurate models of deformable materials. While more data-driven approaches have become more popular for finding controllers that learn to do a single task, there is still a strong motivation for a model-based method that could be used to solve a large variety of optimization problems. Towards this end, we introduced compliant, position-based dynamics (XPBD) to model rope-like objects. Using geometric constraints, the model can represent the coupling of shear/stretch and bend/twist effects. Of crucial importance is that our formulation is differentiable, which can solve parameter estimation problems and improve the matching of rope physics to real-life scenarios (i.e., the real-to-sim problem). For the generality of rope-like objects, two different solvers are proposed to handle the inextensible and extensible effects of varied material stiffness for the rope. We demonstrate our framework's robustness and accuracy on real-to-sim experimental setups using the Baxter robot and the da Vinci research kit (DVRK). Our work leads to a new path for robotic manipulation of the deformable rope-like object taking advantage of the ready-to-use gradients.
Simulation modeling of robots, objects, and environments is the backbone for all model-based control and learning. It is leveraged broadly across dynamic programming and model-predictive control, as well as data generation for imitation, transfer, and reinforcement learning. In addition to fidelity, key features of models in these control and learning contexts are speed, stability, and native differentiability. However, many popular simulation platforms for robotics today lack at least one of the features above. More recently, position-based dynamics (PBD) has become a very popular simulation tool for modeling complex scenes of rigid and non-rigid object interactions, due to its speed and stability, and is starting to gain significant interest in robotics for its potential use in model-based control and learning. Thus, in this paper, we present a mathematical formulation for coupling position-based dynamics (PBD) simulation and optimal robot design, model-based motion control and system identification. Our framework breaks down PBD definitions and derivations for various types of joint-based articulated rigid bodies. We present a back-propagation method with automatic differentiation, which can integrate both positional and angular geometric constraints. Our framework can critically provide the native gradient information and perform gradient-based optimization tasks. We also propose articulated joint model representations and simulation workflow for our differentiable framework. We demonstrate the capability of the framework in efficient optimal robot design, accurate trajectory torque estimation and supporting spring stiffness estimation, where we achieve minor errors. We also implement impedance control in real robots to demonstrate the potential of our differentiable framework in human-in-the-loop applications.
A crucial difference between single- and multi-document summarization is how salient content manifests itself in the document(s). While such content may appear at the beginning of a single document, essential information is frequently reiterated in a set of documents related to a particular topic, resulting in an endorsement effect that increases information salience. In this paper, we model the cross-document endorsement effect and its utilization in multiple document summarization. Our method generates a synopsis from each document, which serves as an endorser to identify salient content from other documents. Strongly endorsed text segments are used to enrich a neural encoder-decoder model to consolidate them into an abstractive summary. The method has a great potential to learn from fewer examples to identify salient content, which alleviates the need for costly retraining when the set of documents is dynamically adjusted. Through extensive experiments on benchmark multi-document summarization datasets, we demonstrate the effectiveness of our proposed method over strong published baselines. Finally, we shed light on future research directions and discuss broader challenges of this task using a case study.
With the explosive growth of livestream broadcasting, there is an urgent need for new summarization technology that enables us to create a preview of streamed content and tap into this wealth of knowledge. However, the problem is nontrivial due to the informal nature of spoken language. Further, there has been a shortage of annotated datasets that are necessary for transcript summarization. In this paper, we present StreamHover, a framework for annotating and summarizing livestream transcripts. With a total of over 500 hours of videos annotated with both extractive and abstractive summaries, our benchmark dataset is significantly larger than currently existing annotated corpora. We explore a neural extractive summarization model that leverages vector-quantized variational autoencoder to learn latent vector representations of spoken utterances and identify salient utterances from the transcripts to form summaries. We show that our model generalizes better and improves performance over strong baselines. The results of this study provide an avenue for future research to improve summarization solutions for efficient browsing of livestreams.
In this paper, we propose an Omni-perception Pre-Trainer (OPT) for cross-modal understanding and generation, by jointly modeling visual, text and audio resources. OPT is constructed in an encoder-decoder framework, including three single-modal encoders to generate token-based embeddings for each modality, a cross-modal encoder to encode the correlations among the three modalities, and two cross-modal decoders to generate text and image respectively. For the OPT's pre-training, we design a multi-task pretext learning scheme to model multi-modal resources from three different data granularities, \ie, token-, modality-, and sample-level modeling, through which OPT learns to align and translate among different modalities. The pre-training task is carried out on a large amount of image-text-audio triplets from Open Images. Experimental results show that OPT can learn strong image-text-audio multi-modal representations and achieve promising results on a variety of cross-modal understanding and generation tasks.
Modeling of non-rigid object launching and manipulation is complex considering the wide range of dynamics affecting trajectory, many of which may be unknown. Using physics models can be inaccurate because they cannot account for unknown factors and the effects of the deformation of the object as it is launched; moreover, deriving force coefficients for these models is not possible without extensive experimental testing. Recently, advancements in data-powered artificial intelligence methods have allowed learnable models and systems to emerge. It is desirable to train a model for launch prediction on a robot, as deep neural networks can account for immeasurable dynamics. However, the inability to collect large amounts of experimental data decreases performance of deep neural networks. Through estimating force coefficients, the accepted physics models can be leveraged to produce adequate supplemental data to artificially increase the size of the training set, yielding improved neural networks. In this paper, we introduce a new framework for algorithmic estimation of force coefficients for non-rigid object launching, which can be generalized to other domains, in order to generate large datasets. We implement a novel training algorithm and objective for our deep neural network to accurately model launch trajectory of non-rigid objects and predict whether they will hit a series of targets. Our experimental results demonstrate the effectiveness of using simulated data from force coefficient estimation and shows the importance of simulated data for training an effective neural network.
Meeting minutes record any subject matters discussed, decisions reached and actions taken at meetings. The importance of minuting cannot be overemphasized in a time when a significant number of meetings take place in the virtual space. In this paper, we present a sliding window approach to automatic generation of meeting minutes. It aims to tackle issues associated with the nature of spoken text, including lengthy transcripts and lack of document structure, which make it difficult to identify salient content to be included in the meeting minutes. Our approach combines a sliding window and a neural abstractive summarizer to navigate through the transcripts to find salient content. The approach is evaluated on transcripts of natural meeting conversations, where we compare results obtained for human transcripts and two versions of automatic transcripts and discuss how and to what extent the summarizer succeeds at capturing salient content.
Monocular 3D object detection is drawing increasing attention from the community as it enables cars to perceive the world in 3D with a single camera. However, monocular 3D detection currently struggles with extremely lower detection rates compared to LiDAR-based methods, limiting its applications. The poor accuracy is mainly caused by the absence of accurate depth cues due to the ill-posed nature of monocular imagery. LiDAR point clouds, which provide accurate depth measurement, can offer beneficial information for the training of monocular methods. Prior works only use LiDAR point clouds to train a depth estimator. This implicit way does not fully utilize LiDAR point clouds, consequently leading to suboptimal performances. To effectively take advantage of LiDAR point clouds, in this paper we propose a general, simple yet effective framework for monocular methods. Specifically, we use LiDAR point clouds to directly guide the training of monocular 3D detectors, allowing them to learn desired objectives meanwhile eliminating the extra annotation cost. Thanks to the general design, our method can be plugged into any monocular 3D detection method, significantly boosting the performance. In conclusion, we take the first place on KITTI monocular 3D detection benchmark and increase the BEV/3D AP from 11.88/8.65 to 22.06/16.80 on the hard setting for the prior state-of-the-art method. The code will be made publicly available soon.
Image-only and pseudo-LiDAR representations are commonly used for monocular 3D object detection. However, methods based on them have shortcomings of either not well capturing the spatial relationships in neighbored image pixels or being hard to handle the noisy nature of the monocular pseudo-LiDAR point cloud. To overcome these issues, in this paper we propose a novel object-centric voxel representation tailored for monocular 3D object detection. Specifically, voxels are built on each object proposal, and their sizes are adaptively determined by the 3D spatial distribution of the points, allowing the noisy point cloud to be organized effectively within a voxel grid. This representation is proved to be able to locate the object in 3D space accurately. Furthermore, prior works would like to estimate the orientation via deep features extracted from an entire image or a noisy point cloud. By contrast, we argue that the local RoI information from the object image patch alone with a proper resizing scheme is a better input as it provides complete semantic clues meanwhile excludes irrelevant interferences. Besides, we decompose the confidence mechanism in monocular 3D object detection by considering the relationship between 3D objects and the associated 2D boxes. Evaluated on KITTI, our method outperforms state-of-the-art methods by a large margin. The code will be made publicly available soon.
We propose a new approach to generate multiple variants of the target summary with diverse content and varying lengths, then score and select admissible ones according to users' needs. Abstractive summarizers trained on single reference summaries may struggle to produce outputs that achieve multiple desirable properties, i.e., capturing the most important information, being faithful to the original, grammatical and fluent. In this paper, we propose a two-staged strategy to generate a diverse set of candidate summaries from the source text in stage one, then score and select admissible ones in stage two. Importantly, our generator gives a precise control over the length of the summary, which is especially well-suited when space is limited. Our selectors are designed to predict the optimal summary length and put special emphasis on faithfulness to the original text. Both stages can be effectively trained, optimized and evaluated. Our experiments on benchmark summarization datasets suggest that this paradigm can achieve state-of-the-art performance.