When humans perform contact-rich manipulation tasks, customized tools are often necessary and play an important role in simplifying the task. For instance, in our daily life, we use various utensils for handling food, such as knives, forks and spoons. Similarly, customized tools for robots may enable them to more easily perform a variety of tasks. Here, we present an end-to-end framework to automatically learn tool morphology for contact-rich manipulation tasks by leveraging differentiable physics simulators. Previous work approached this problem by introducing manually constructed priors that required detailed specification of object 3D model, grasp pose and task description to facilitate the search or optimization. In our approach, we instead only need to define the objective with respect to the task performance and enable learning a robust morphology by randomizing the task variations. The optimization is made tractable by casting this as a continual learning problem. We demonstrate the effectiveness of our method for designing new tools in several scenarios such as winding ropes, flipping a box and pushing peas onto a scoop in simulation. We also validate that the shapes discovered by our method help real robots succeed in these scenarios.

可区分的仿真是用于基于快速梯度的策略优化和系统识别的有前途的工具包。但是，现有的可区分仿真方法在很大程度上已经解决了获得平滑梯度相对容易的方案，例如具有光滑动力学的系统。在这项工作中，我们研究了可区分的模拟所面临的挑战，当时单个下降不可行，这通常是全球最佳的，这通常是接触率丰富的方案中的问题。我们分析包含刚体和可变形物体的各种情况的优化景观。在具有高度可变形的物体和流体的动态环境中，可区分的模拟器在空间的某些地方生产具有有用梯度的坚固景观。我们提出了一种将贝叶斯优化与半本地“飞跃”相结合的方法，以获得可以有效使用梯度的全局搜索方法，同时还可以在具有嘈杂梯度的地区保持稳健的性能。我们表明，我们的方法在模拟中的一组实验集上优于几个基于梯度和无梯度的基线，并且还使用具有真实机器人和变形物的实验验证该方法。视频和补充材料可从https://tinyurl.com/globdiff获得

可变形的物体操纵仍然是机器人研究中的具有挑战性的任务。用于参数推断和状态估计的传统技术通常依赖于状态空间的精确定义及其动态。虽然这适用于刚性物体和机器人状态，但定义可变形物体的状态空间并如何及时演变。在这项工作中，我们构成了作为用模拟器定义的概率推断任务的可变形对象的物理参数的问题。我们提出了一种用于通过技术从图像序列提取状态信息的新方法，以将可变形对象作为分布嵌入的状态提取。这允许以原则的方式将噪声状态观察直接进入基于现代贝叶斯模拟的推理工具。我们的实验证实，我们可以估计物理性质的后部分布，例如高可变形物体的弹性，摩擦和尺度，例如布和绳索。总的来说，我们的方法解决了概率的实际问题，并有助于更好地代表可变形对象状态的演变。

Pairwise compatibility measure (CM) is a key component in solving the jigsaw puzzle problem (JPP) and many of its recently proposed variants. With the rapid rise of deep neural networks (DNNs), a trade-off between performance (i.e., accuracy) and computational efficiency has become a very significant issue. Whereas an end-to-end DNN-based CM model exhibits high performance, it becomes virtually infeasible on very large puzzles, due to its highly intensive computation. On the other hand, exploiting the concept of embeddings to alleviate significantly the computational efficiency, has resulted in degraded performance, according to recent studies. This paper derives an advanced CM model (based on modified embeddings and a new loss function, called hard batch triplet loss) for closing the above gap between speed and accuracy; namely a CM model that achieves SOTA results in terms of performance and efficiency combined. We evaluated our newly derived CM on three commonly used datasets, and obtained a reconstruction improvement of 5.8% and 19.5% for so-called Type-1 and Type-2 problem variants, respectively, compared to best known results due to previous CMs.

逆文本归一化（ITN）是自动语音识别（ASR）中必不可少的后处理步骤。它将数字，日期，缩写和其他符号类别从ASR产生的口头形式转换为其书面形式。人们可以将ITN视为机器翻译任务，并使用神经序列到序列模型来解决它。不幸的是，这种神经模型容易产生可能导致不可接受的错误的幻觉。为了减轻此问题，我们提出了一个单个令牌分类器模型，将ITN视为标记任务。该模型将替换片段分配给每个输入令牌，或将其标记为删除或复制而无需更改。我们提出了基于ITN示例的粒状对齐方式的数据集准备方法。提出的模型不太容易出现幻觉错误。该模型在Google文本归一化数据集上进行了培训，并在英语和俄罗斯测试集上实现了最先进的句子精度。标签和输入单词之间的一对一对应关系可改善模型预测的解释性，简化调试并允许后处理更正。该模型比序列到序列模型更简单，并且在生产设置中更易于优化。准备数据集的模型和代码作为NEMO项目的一部分发布。

This paper introduces the novel CNN-based encoder Twin Embedding Network (TEN), for the jigsaw puzzle problem (JPP), which represents a puzzle piece with respect to its boundary in a latent embedding space. Combining this latent representation with a simple distance measure, we demonstrate improved accuracy levels of our newly proposed pairwise compatibility measure (CM), compared to that of various classical methods, for degraded puzzles with eroded tile boundaries. We focus on this problem instance for our case study, as it serves as an appropriate testbed for real-world scenarios. Specifically, we demonstrated an improvement of up to 8.5% and 16.8% in reconstruction accuracy, for so-called Type-1 and Type-2 problem variants, respectively. Furthermore, we also demonstrated that TEN is faster by a few orders of magnitude, on average, than a typical deep neural network (NN) model, i.e., it is as fast as the classical methods. In this regard, the paper makes a significant first attempt at bridging the gap between the relatively low accuracy (of classical methods and the intensive computational complexity (of NN models), for practical, real-world puzzle-like problems.