In this work, we introduce the Virtual In-Hand Eye Transformer (VIHE), a novel method designed to enhance 3D manipulation capabilities through action-aware view rendering. VIHE autoregressively refines actions in multiple stages by conditioning on rendered views posed from action predictions in the earlier stages. These virtual in-hand views provide a strong inductive bias for effectively recognizing the correct pose for the hand, especially for challenging high-precision tasks such as peg insertion. On 18 manipulation tasks in RLBench simulated environments, VIHE achieves a new state-of-the-art, with a 12% absolute improvement, increasing from 65% to 77% over the existing state-of-the-art model using 100 demonstrations per task. In real-world scenarios, VIHE can learn manipulation tasks with just a handful of demonstrations, highlighting its practical utility. Videos and code implementation can be found at our project site: https://vihe-3d.github.io.

Recent advances in latent space dynamics model from pixels show promising progress in vision-based model predictive control (MPC). However, executing MPC in real time can be challenging due to its intensive computational cost in each timestep. We propose to introduce additional learning objectives to enforce that the learned latent space is proportional derivative controllable. In execution time, the simple PD-controller can be applied directly to the latent space encoded from pixels, to produce simple and effective control to systems with visual observations. We show that our method outperforms baseline methods to produce robust goal reaching and trajectory tracking in various environments.

Prior work demonstrated the ability of machine learning to automatically recognize surgical workflow steps from videos. However, these studies focused on only a single type of procedure. In this work, we analyze, for the first time, surgical step recognition on four different laparoscopic surgeries: Cholecystectomy, Right Hemicolectomy, Sleeve Gastrectomy, and Appendectomy. Inspired by the traditional apprenticeship model, in which surgical training is based on the Halstedian method, we paraphrase the "see one, do one, teach one" approach for the surgical intelligence domain as "train one, classify one, teach one". In machine learning, this approach is often referred to as transfer learning. To analyze the impact of transfer learning across different laparoscopic procedures, we explore various time-series architectures and examine their performance on each target domain. We introduce a new architecture, the Time-Series Adaptation Network (TSAN), an architecture optimized for transfer learning of surgical step recognition, and we show how TSAN can be pre-trained using self-supervised learning on a Sequence Sorting task. Such pre-training enables TSAN to learn workflow steps of a new laparoscopic procedure type from only a small number of labeled samples from the target procedure. Our proposed architecture leads to better performance compared to other possible architectures, reaching over 90% accuracy when transferring from laparoscopic Cholecystectomy to the other three procedure types.

A fundamental challenge in retinal surgery is safely navigating a surgical tool to a desired goal position on the retinal surface while avoiding damage to surrounding tissues, a procedure that typically requires tens-of-microns accuracy. In practice, the surgeon relies on depth-estimation skills to localize the tool-tip with respect to the retina in order to perform the tool-navigation task, which can be prone to human error. To alleviate such uncertainty, prior work has introduced ways to assist the surgeon by estimating the tool-tip distance to the retina and providing haptic or auditory feedback. However, automating the tool-navigation task itself remains unsolved and largely unexplored. Such a capability, if reliably automated, could serve as a building block to streamline complex procedures and reduce the chance for tissue damage. Towards this end, we propose to automate the tool-navigation task by learning to mimic expert demonstrations of the task. Specifically, a deep network is trained to imitate expert trajectories toward various locations on the retina based on recorded visual servoing to a given goal specified by the user. The proposed autonomous navigation system is evaluated in simulation and in physical experiments using a silicone eye phantom. We show that the network can reliably navigate a needle surgical tool to various desired locations within 137 microns accuracy in physical experiments and 94 microns in simulation on average, and generalizes well to unseen situations such as in the presence of auxiliary surgical tools, variable eye backgrounds, and brightness conditions.

CXRs are a crucial and extraordinarily common diagnostic tool, leading to heavy research for CAD solutions. However, both high classification accuracy and meaningful model predictions that respect and incorporate clinical taxonomies are crucial for CAD usability. To this end, we present a deep HMLC approach for CXR CAD. Different than other hierarchical systems, we show that first training the network to model conditional probability directly and then refining it with unconditional probabilities is key in boosting performance. In addition, we also formulate a numerically stable cross-entropy loss function for unconditional probabilities that provides concrete performance improvements. Finally, we demonstrate that HMLC can be an effective means to manage missing or incomplete labels. To the best of our knowledge, we are the first to apply HMLC to medical imaging CAD. We extensively evaluate our approach on detecting abnormality labels from the CXR arm of the PLCO dataset, which comprises over $198,000$ manually annotated CXRs. When using complete labels, we report a mean AUC of 0.887, the highest yet reported for this dataset. These results are supported by ancillary experiments on the PadChest dataset, where we also report significant improvements, 1.2% and 4.1% in AUC and AP, respectively over strong "flat" classifiers. Finally, we demonstrate that our HMLC approach can much better handle incompletely labelled data. These performance improvements, combined with the inherent usefulness of taxonomic predictions, indicate that our approach represents a useful step forward for CXR CAD.

"Good Robot!": Efficient Reinforcement Learning for Multi-Step Visual T asks via Reward Shaping Andrew Hundt 1, Benjamin Killeen 1, Heeyeon Kwon 1, Chris Paxton 2, and Gregory D. Hager 1 Abstract -- In order to learn effectively, robots must be able to extract the intangible context by which task progress and mistakes are defined. In the domain of reinforcement learning, much of this information is provided by the reward function. Hence, reward shaping is a necessary part of how we can achieve state-of-the-art results on complex, multi-step tasks. However, comparatively little work has examined how reward shaping should be done so that it captures task context, particularly in scenarios where the task is long-horizon and failure is highly consequential. Our Schedule for Positive T ask (SPOT) reward trains our Efficient Visual T ask (EVT) model to solve problems that require an understanding of both task context and workspace constraints of multi-step block arrangement tasks. In simulation EVT can completely clear adversarial arrangements of objects by pushing and grasping in 99% of cases vs an 82% baseline in prior work. For random arrangements EVT clears 100% of test cases at 86% action efficiency vs 61% efficiency in prior work. EVT SPOT is also able to demonstrate context understanding and complete stacks in 74% of trials compared to a baseline of 5% with EVT alone.

Neural Architecture Search (NAS) has been a source of dramatic improvements in neural network design, with recent results meeting or exceeding the performance of hand-tuned architectures. However, our understanding of how to represent the search space for neural net architectures and how to search that space efficiently are both still in their infancy. We have performed an in-depth analysis to identify limitations in a widely used search space and a recent architecture search method, Differentiable Architecture Search (DARTS). These findings led us to introduce novel network blocks with a more general, balanced, and consistent design; a better-optimized Cosine Power Annealing learning rate schedule; and other improvements. Our resulting sharpDARTS search is 50% faster with a 20-30% relative improvement in final model error on CIFAR-10 when compared to DARTS. Our best single model run has 1.93% (1.98+/-0.07) validation error on CIFAR-10 and 5.5% error (5.8+/-0.3) on the recently released CIFAR-10.1 test set. To our knowledge, both are state of the art for models of similar size. This model also generalizes competitively to ImageNet at 25.1% top-1 (7.8% top-5) error. We found improvements for existing search spaces but does DARTS generalize to new domains? We propose Differentiable Hyperparameter Grid Search and the HyperCuboid search space, which are representations designed to leverage DARTS for more general parameter optimization. Here we find that DARTS fails to generalize when compared against a human's one shot choice of models. We look back to the DARTS and sharpDARTS search spaces to understand why, and an ablation study reveals an unusual generalization gap. We finally propose Max-W regularization to solve this problem, which proves significantly better than the handmade design. Code will be made available.

A robot can now grasp an object more effectively than ever before, but once it has the object what happens next? We show that a mild relaxation of the task and workspace constraints implicit in existing object grasping datasets can cause neural network based grasping algorithms to fail on even a simple block stacking task when executed under more realistic circumstances. To address this, we introduce the JHU CoSTAR Block Stacking Dataset (BSD), where a robot interacts with 5.1 cm colored blocks to complete an order-fulfillment style block stacking task. It contains dynamic scenes and real time-series data in a less constrained environment than comparable datasets. There are nearly 12,000 stacking attempts and over 2 million frames of real data. We discuss the ways in which this dataset provides a valuable resource for a broad range of other topics of investigation. We find that hand-designed neural networks that work on prior datasets do not generalize to this task. Thus, to establish a baseline for this dataset, we demonstrate an automated search of neural network based models using a novel multiple-input HyperTree MetaModel, and find a final model which makes reasonable 3D pose predictions for grasping and stacking on our dataset. The CoSTAR BSD, code, and instructions are available at https://sites.google.com/site/costardataset.

INIMALLY invasiveprocedures in the head and neck, e. g. functional endoscopic sinus surgery, typically employ surgical navigation systems to provide surgeons with additional anatomical and positional information. This helps them avoid critical structures, such as the brain, eyes, and major arteries, that are spatially close to the sinus cavities and must not be disturbed during surgery. Computer vision-based navigation systems that rely on the intra-operative endoscopic video stream and do not introduce additional hardware are both easy to integrate into clinical workflow and cost-effective. Such systems generally require registration of preoperative data, such as CT scans or statistical models, to the intraoperative videodata [1], [2], [3], [4]. This registration must be highly accurate in order to guarantee reliable performance of the navigation system. To enable an accurate registration, a feature-based video-CT registration algorithm requires accurate andsufficiently dense intra-operative 3D reconstructions of the anatomy from endoscopic videos.

In this paper, we propose a novel exemplar-based approach to extract dynamic foreground regions from a changing background within a collection of images or a video sequence. By using image segmentation as a pre-processing step, we convert this traditional pixel-wise labeling problem into a lower-dimensional supervised, binary labeling procedure on image segments. Our approach consists of three steps. First, a set of random image patches are spatially and adaptively sampled within each segment. Second, these sets of extracted samples are formed into two "bags of patches" to model the foreground/background appearance, respectively.