ICRA 2017 论文列表

3-D perception of scenes with specular surfaces is still challenging for robotics applications in urban areas, for both active and passive range sensors; there is a need for improved solutions that work without artificial illumination over a wide range of distances. The advent of cameras with microgrid polarization filter arrays, which allow acquiring four orientations of linearly polarized images simultaneously, has potential to make the use of polarization information in 3-D perception more practical. It is well-known that polarization can provide information about the orientation of specular surfaces; however, prior work with polarization for 3-D perception has had several limitations. We present the first unified formulation of depth perception with stereo and polarization by extending previous energy minimization formulations to include surface orientation constraints computed from the polarization channels. We apply an existing quadratic pseudo-boolean optimization (QPBO) method to approximate the optimal depth map. We use synthetic and real indoor/outdoor images to demonstrate that the new method achieves better results than prior methods, with fewer assumptions and limitations.

This paper describes the outcome of a pilot study of deploying humanoid social robots as a teaching aid in preschool classroom. Sixteen K2 students each from two pre-schools were recruited to test commercially available robots, Pepper and Nao. The robots were assigned to different schools and were programmed to deliver 6 lessons over a span of 3 months. To assess the deployment and the performance of the children, we used Likert-scale based survey to record our observations. One of the survey forms used was TEPI, which measures the performance of the children's behavior based on three major criteria. The TEPI data gathered suggests that during the lessons children display desirable behavior such as critical thinking, imagination and creativity, and social interaction and independence. We also observed classroom atmosphere, classroom management, and class behavior during the lessons. Data from the survey highlights the potential benefits and the challenges of deploying robots as a teaching aid in a real classroom setting.

Flexible endoscopes are still the gold standard in most natural orifice translumenal endoscopic surgery (NOTES) procedures; however their flexibility (necessary for navigating through the GI tract) limits their capabilities in terms of distal manipulation and stability. We propose a deployable endoscopic add-on aimed at locally counteracting forces applied at the tip of an endoscope. We analyze different designs: a fully soft version and two hybrid soft-folded versions. The hybrid designs exploit either an inextensible structure pressurized by a soft actuator or the stiffness provided by the unfolded “magic cube” origami structure. We focus on the fabrication and experimental characterization of the proposed structures and present some preliminary designs and integration strategies to mount them on top of current flexible endoscopes.

We consider the problem of computing shortest paths in a dense motion-planning roadmap G. We assume that n, the number of vertices of G, is very large. Thus, using any path-planning algorithm that directly searches G, running in O(VlogV + E) ≈ O(n2) time, becomes unacceptably expensive. We are therefore interested in anytime search to obtain successively shorter feasible paths and converge to the shortest path in G. Our key insight is to provide existing path-planning algorithms with a sequence of increasingly dense subgraphs of G. We study the space of all (r-disk) subgraphs of G. We then formulate and present two densification strategies for traversing this space which exhibit complementary properties with respect to problem difficulty. This inspires a third, hybrid strategy which has favourable properties regardless of problem difficulty. This general approach is then demonstrated and analyzed using the specific case where a low-dispersion deterministic sequence is used to generate the samples used for G. Finally we empirically evaluate the performance of our strategies for random scenarios in ℝ2 and ℝ4 and on manipulation planning problems for a 7 DOF robot arm, and validate our analysis.

This paper presents an efficient framework for dense 3D scene reconstruction using input from a moving monocular camera. Visual SLAM (Simultaneous Localisation and Mapping) approaches based solely on geometric methods have proven to be quite capable of accurately tracking the pose of a moving camera and simultaneously building a map of the environment in real-time. However, most of them suffer from the 3D map being too sparse for practical use. The missing points in the generated map correspond mainly to areas lacking texture in the input images, and dense mapping systems often rely on hand-crafted priors like piecewise-planarity or piecewise-smooth depth. These priors do not always provide the required level of scene understanding to accurately fill the map. On the other hand, Convolutional Neural Networks (CNNs) have had great success in extracting high-level information from images and regressing pixel-wise surface normals, semantics, and even depth. In this work we leverage this high-level scene context learned by a deep CNN in the form of a surface normal prior. We show, in particular, that using the surface normal prior leads to better reconstructions than the weaker smoothness prior.

Convolutional Neural Networks (CNNs) have been applied to camera relocalization, which is to infer the pose of the camera given a single monocular image. However, there are still many open problems for camera relocalization with CNNs. We delve into the CNNs for camera relocalization. First, a variant of Euler angles named Euler6 is proposed to represent orientation. Then a data augmentation method named pose synthesis is designed to reduce spsarsity of poses in the whole pose space to cope with overfitting in training. Third, a multi-task CNN named BranchNet is proposed to deal with the complex coupling of orientation and translation. The network consists of several shared convolutional layers and splits into two branches which predict orientation and translation, respectively. Experiments on the 7Scenes dataset show that incorporating these techniques one by one into an existing model PoseNet always leads to better results. Together these techniques reduce the orientation error by 15.9% and the translation error by 38.3% compared to the state-of-the-art model Bayesian PoseNet. We implement BranchNet on an Intel NUC mobile platform and reach a speed of 43 fps, which meets the real-time requirement of many robotic applications.

A fundamental robot perception task is that of identifying and estimating the poses of objects with known 3D models in RGB-D data. While feature-based and discriminative approaches have been traditionally used for this task, recent work on deliberative approaches such as PERCH and D2P have shown improved robustness in handling scenes with severe inter-object occlusions. These deliberative approaches work by treating multi-object pose estimation as a combinatorial search over the space of possible rendered scenes of the objects, thereby inherently being able to predict and account for occlusions. However, these methods have so far been restricted to scenes comprising only of known objects, and have been unable to handle extraneous clutter — a common occurrence in many real-world settings. This work significantly increases the practical relevance of deliberative perception methods by developing a formulation that: i) accounts for extraneous unmodeled clutter in scenes, and ii) provides object pose uncertainty estimates. Our algorithm is complete and provides bounded suboptimality guarantees for the cost function chosen to be optimized. Empirically, we demonstrate successful object recognition and uncertainty-aware localization in challenging scenes with unmodeled clutter, where previous deliberative methods perform unsatisfactorily. In addition, this work was used as part of the perception system by Carnegie Mellon University's Team HARP in the 2016 Amazon Picking Challenge.

This paper studies monocular visual odometry (VO) problem. Most of existing VO algorithms are developed under a standard pipeline including feature extraction, feature matching, motion estimation, local optimisation, etc. Although some of them have demonstrated superior performance, they usually need to be carefully designed and specifically fine-tuned to work well in different environments. Some prior knowledge is also required to recover an absolute scale for monocular VO. This paper presents a novel end-to-end framework for monocular VO by using deep Recurrent Convolutional Neural Networks (RCNNs). Since it is trained and deployed in an end-to-end manner, it infers poses directly from a sequence of raw RGB images (videos) without adopting any module in the conventional VO pipeline. Based on the RCNNs, it not only automatically learns effective feature representation for the VO problem through Convolutional Neural Networks, but also implicitly models sequential dynamics and relations using deep Recurrent Neural Networks. Extensive experiments on the KITTI VO dataset show competitive performance to state-of-the-art methods, verifying that the end-to-end Deep Learning technique can be a viable complement to the traditional VO systems.

A key challenge in scaling up robot learning to many skills and environments is removing the need for human supervision, so that robots can collect their own data and improve their own performance without being limited by the cost of requesting human feedback. Model-based reinforcement learning holds the promise of enabling an agent to learn to predict the effects of its actions, which could provide flexible predictive models for a wide range of tasks and environments, without detailed human supervision. We develop a method for combining deep action-conditioned video prediction models with model-predictive control that uses entirely unlabeled training data. Our approach does not require a calibrated camera, an instrumented training set-up, nor precise sensing and actuation. Our results show that our method enables a real robot to perform nonprehensile manipulation - pushing objects - and can handle novel objects not seen during training.

In this paper, we present a visual learning framework to retrieve a 3D model and estimate its pose from a single image. To increase the quantity and quality of training data, we define our simulation space in the near infrared (NIR) band, and utilize the quasi-Monte Carlo (MC) method for scalable photorealistic rendering of manufactured components. Two types of convolutional neural network (CNN) architectures are trained over these synthetic data and a relatively small amount of real data. The first CNN model seeks the most discriminative information and uses it to classify industrial components with fine-grained shape attributes. Once a 3D model is identified, one of the category-specific CNNs is tested for pose regression in the second phase. The mixed data for learning object categories is useful in domain adaptation and attention mechanism in our system. We validate our data-driven method with 88 component models, and the experimental results are qualitatively demonstrated. Also, the CNNs trained with various conditions of mixed data are quantitatively analyzed.

Tensegrity robots, composed of rigid rods connected by elastic cables, have a number of unique properties that make them appealing for use as planetary exploration rovers. However, control of tensegrity robots remains a difficult problem due to their unusual structures and complex dynamics. In this work, we show how locomotion gaits can be learned automatically using a novel extension of mirror descent guided policy search (MDGPS) applied to periodic locomotion movements, and we demonstrate the effectiveness of our approach on tensegrity robot locomotion. We evaluate our method with real-world and simulated experiments on the SUPERball tensegrity robot, showing that the learned policies generalize to changes in system parameters, unreliable sensor measurements, and variation in environmental conditions, including varied terrains and a range of different gravities. Our experiments demonstrate that our method not only learns fast, power-efficient feedback policies for rolling gaits, but that these policies can succeed with only the limited onboard sensing provided by SUPERball's accelerometers. We compare the learned feedback policies to learned open-loop policies and hand-engineered controllers, and demonstrate that the learned policy enables the first continuous, reliable locomotion gait for the real SUPERball robot. Our code and supplementary material is available from http://rll.berkeley.edu/drl_tensegrity.

Reinforcement learning holds the promise of enabling autonomous robots to learn large repertoires of behavioral skills with minimal human intervention. However, robotic applications of reinforcement learning often compromise the autonomy of the learning process in favor of achieving training times that are practical for real physical systems. This typically involves introducing hand-engineered policy representations and human-supplied demonstrations. Deep reinforcement learning alleviates this limitation by training general-purpose neural network policies, but applications of direct deep reinforcement learning algorithms have so far been restricted to simulated settings and relatively simple tasks, due to their apparent high sample complexity. In this paper, we demonstrate that a recent deep reinforcement learning algorithm based on off-policy training of deep Q-functions can scale to complex 3D manipulation tasks and can learn deep neural network policies efficiently enough to train on real physical robots. We demonstrate that the training times can be further reduced by parallelizing the algorithm across multiple robots which pool their policy updates asynchronously. Our experimental evaluation shows that our method can learn a variety of 3D manipulation skills in simulation and a complex door opening skill on real robots without any prior demonstrations or manually designed representations.

Trajectory tracking control for quadrotors is important for applications ranging from surveying and inspection, to film making. However, designing and tuning classical controllers, such as proportional-integral-derivative (PID) controllers, to achieve high tracking precision can be time-consuming and difficult, due to hidden dynamics and other non-idealities. The Deep Neural Network (DNN), with its superior capability of approximating abstract, nonlinear functions, proposes a novel approach for enhancing trajectory tracking control. This paper presents a DNN-based algorithm as an add-on module that improves the tracking performance of a classical feedback controller. Given a desired trajectory, the DNNs provide a tailored reference input to the controller based on their gained experience. The input aims to achieve a unity map between the desired and the output trajectory. The motivation for this work is an interactive “fly-as-you-draw” application, in which a user draws a trajectory on a mobile device, and a quadrotor instantly flies that trajectory with the DNN-enhanced control system. Experimental results demonstrate that the proposed approach improves the tracking precision for user-drawn trajectories after the DNNs are trained on selected periodic trajectories, suggesting the method's potential in real-world applications. Tracking errors are reduced by around 40-50% for both training and testing trajectories from users, highlighting the DNNs' capability of generalizing knowledge.

A robot operating in a real-world environment needs to perform reasoning over a variety of sensor modalities such as vision, language and motion trajectories. However, it is extremely challenging to manually design features relating such disparate modalities. In this work, we introduce an algorithm that learns to embed point-cloud, natural language, and manipulation trajectory data into a shared embedding space with a deep neural network. To learn semantically meaningful spaces throughout our network, we use a loss-based margin to bring embeddings of relevant pairs closer together while driving less-relevant cases from different modalities further apart. We use this both to pre-train its lower layers and fine-tune our final embedding space, leading to a more robust representation. We test our algorithm on the task of manipulating novel objects and appliances based on prior experience with other objects. On a large dataset, we achieve significant improvements in both accuracy and inference time over the previous state of the art. We also perform end-to-end experiments on a PR2 robot utilizing our learned embedding space.

The success of deep learning techniques in the computer vision domain has triggered a range of initial investigations into their utility for visual place recognition, all using generic features from networks that were trained for other types of recognition tasks. In this paper, we train, at large scale, two CNN architectures for the specific place recognition task and employ a multi-scale feature encoding method to generate condition- and viewpoint-invariant features. To enable this training to occur, we have developed a massive Specific PlacEs Dataset (SPED) with hundreds of examples of place appearance change at thousands of different places, as opposed to the semantic place type datasets currently available. This new dataset enables us to set up a training regime that interprets place recognition as a classification problem. We comprehensively evaluate our trained networks on several challenging benchmark place recognition datasets and demonstrate that they achieve an average 10% increase in performance over other place recognition algorithms and pre-trained CNNs. By analyzing the network responses and their differences from pre-trained networks, we provide insights into what a network learns when training for place recognition, and what these results signify for future research in this area.

An accurate and reliable image based fruit detection system is critical for supporting higher level agriculture tasks such as yield mapping and robotic harvesting. This paper presents the use of a state-of-the-art object detection framework, Faster R-CNN, in the context of fruit detection in orchards, including mangoes, almonds and apples. Ablation studies are presented to better understand the practical deployment of the detection network, including how much training data is required to capture variability in the dataset. Data augmentation techniques are shown to yield significant performance gains, resulting in a greater than two-fold reduction in the number of training images required. In contrast, transferring knowledge between orchards contributed to negligible performance gain over initialising the Deep Convolutional Neural Network directly from ImageNet features. Finally, to operate over orchard data containing between 100-1000 fruit per image, a tiling approach is introduced for the Faster R-CNN framework. The study has resulted in the best yet detection performance for these orchards relative to previous works, with an F1-score of > 0.9 achieved for apples and mangoes.

When limbs are decoupled, we find that trajectory outcomes in mechanical systems subject to unilateral constraints vary differentiably with respect to initial conditions, even as the contact mode sequence varies.

Finding feasible, collision-free paths for multiagent systems can be challenging, particularly in non-communicating scenarios where each agent's intent (e.g. goal) is unobservable to the others. In particular, finding time efficient paths often requires anticipating interaction with neighboring agents, the process of which can be computationally prohibitive. This work presents a decentralized multiagent collision avoidance algorithm based on a novel application of deep reinforcement learning, which effectively offloads the online computation (for predicting interaction patterns) to an offline learning procedure. Specifically, the proposed approach develops a value network that encodes the estimated time to the goal given an agent's joint configuration (positions and velocities) with its neighbors. Use of the value network not only admits efficient (i.e., real-time implementable) queries for finding a collision-free velocity vector, but also considers the uncertainty in the other agents' motion. Simulation results show more than 26% improvement in paths quality (i.e., time to reach the goal) when compared with optimal reciprocal collision avoidance (ORCA), a state-of-the-art collision avoidance strategy.

This paper addresses the motion planning problem for a team of aerial agents under high level goals. We propose a hybrid control strategy that guarantees the accomplishment of each agent's local goal specification, which is given as a temporal logic formula, while guaranteeing inter-agent collision avoidance. In particular, by defining 3-D spheres that bound the agents' volume, we extend previous work on decentralized navigation functions and propose control laws that navigate the agents among predefined regions of interest of the workspace while avoiding collision with each other. This allows us to abstract the motion of the agents as finite transition systems and, by employing standard formal verification techniques, to derive a high-level control algorithm that satisfies the agents' specifications. Simulation and experimental results with quadrotors verify the validity of the proposed method.