ICRA 2019 论文列表

This paper presents a novel quadcopter design that uses passive rotary joints to enable rapid aerial morphing without the use of additional actuators. The normally rigid connections between the arms of the quadcopter and the central body are replaced by sprung hinges that allow for the arms of the quadcopter to fold downward when low thrusts are produced by the propellers, resulting in a reduction of the largest dimension of the vehicle by approximately 50%. The ability of the vehicle to reduce its size during flight allows, e.g., for the traversal of gaps through which a non-morphing quadcopter could not pass. The vehicle is designed such that existing quadcopter controllers and trajectory generation algorithms can be used, provided that some additional constraints on the control inputs are met. The nonlinear dynamics of the system are presented, and design rules are given that minimize transition time between configurations and maximize the available range of control inputs. A method for performing gap traversal maneuvers is proposed and validated experimentally.

Developments of robot arms are countless, but there has been little focus on robot surfaces for the reshaping of a habitable space—especially compliant surfaces. In this paper we introduce a novel, tendon-driven, robot surface comprised of aggregated, overlapping panels organized in a herringbone pattern. The individual 3D-printed panels and their behavior as an aggregation are inspired by the form and behavior of a pinecone. This paper presents our concept, design, and realization of this robot, and compares our prototype to simulations of four physical configurations that are formally distinct and suggestive of how the surface might be applied to habitable, physical space in response to human needs and wants. For the four configurations studied, we found a validating match between prototype and simulations. The paper concludes with a consideration of potential applications for robot surfaces like this one.

Plane-to-plane transitioning has been a significant challenge for climbing robots. To accomplish this, additional actuator or robot module is usually required which significantly increases both size and weight of the robot. This paper presents a two-wheg miniature climbing robot with a novel passive vertical tail component which results in robust transitioning capabilities. The design decision was derived from an indepth force analysis of the climbing robot while performing the transition. The theoretical analysis is verified through a working prototype with robust transitioning capabilities whose performance follows closely the analytical prediction. The climbing robot is able to climb any slope angles, 4-way internal transitions, and 4-way external transitions. This work contributes to the understanding and advancement of the transitioning capabilities and the design of a simple climbing robot, which expands the possibilities of scaling down miniature climbing robot further.

In this paper we address the problem of creating planar caging grasps on objects using simple, underactuated grippers with no sensing or control. Specifically, we examine how changes in mechanical compliance, passive adaptability due to underactuation, and finger phalanx length affect the ability to create caging grasps passively, by altering the free-swing motion of the fingers. We present a simple model for simulating the underactuated hand, develop a metric for quantifying a hand design's caging ability, and perform a design parameter space search to reveal the important design factors influencing passive caging behavior. The results show that both palm width and the interplay between joint spring stiffness and pulley radius ratios play the largest roles in determining caging behavior. The effect of varying design parameters on the caging grasp performance of the hand is discussed, the best resulting design is shown, and a list of principles to guide the design of simple underactuated hands for caging grasps is presented.

In this work, we propose the Depth Generation Network (DGN) to address the problem of dense depth estimation by exploiting the variational method and the deep-learning technique. In particular, we focus on improving the feasibility of depth estimation under complex scenarios given stereo RGB images, where the stereo pairs and/or depth ground-truth captured by real sensors may be deteriorated; the stereo setting parameters may be unavailable or unreliable, hence hamper efforts to establish the correspondence between image pairs via supervision learning or epipolar geometric cues. Instead of relying on real data, we supervise the training of our model using synthetic depth maps generated by the simulator, which deliver complex scenes and reliable data with ease. Two non-trivial challenges, i.e., (i) attaining reasonable amount yet realistic samples for training, and (ii) developing a model that adapts to both synthetic and real scenes arise, whereas in this work we mainly deal with the later one yet leveraging state-of-the-art Falling Things (FAT) dataset to overcome the first. Experiments on FAT and KITTI datasets demonstrate that our model estimates relative dense depth in fine details, potentially generalizable to real scenes without knowing the stereo geometric and optic settings.

In this paper, we present an end-to-end convolutional neural network (CNN) for depth completion. Our network consists of a geometry network and a context network. The geometry network, a single encoder-decoder network, learns to optimize a multi-task loss to generate an initial propagated depth map and a surface normal. The complementary outputs allow it to correctly propagate initial sparse depth points in slanted surfaces. The context network extracts a local and a global feature of an image to compute a bilateral weight, which enables it to preserve edges and fine details in the depth maps. At the end, a final output is produced by multiplying the initially propagated depth map with the bilateral weight. In order to validate the effectiveness and the robustness of our network, we performed extensive ablation studies and compared the results against state-of-the-art CNN-based depth completions, where we showed promising results on various scenes.

In this paper we introduce dentronics as a new emerging application domain for collaborative lightweight robots in the dental context backed up by a user survey supporting the clear need. Specifically, we developed a multi-modal interaction framework, applied this framework to a specific dental use-case, and conducted a preliminary user-study for evaluation. Our results demonstrate usability and feasibility beyond a controlled experimental setup. We conclude that dentronics is indeed within reach given today's technology and deserves further investigation towards clinical use.

In this paper, we present a new version of tactile sensor GelSlim 2.0 with the capability to estimate the contact force distribution in real time. The sensor is vision-based and uses an array of markers to track deformations on a gel pad due to contact. A new hardware design makes the sensor more rugged, parametrically adjusTable AND Improves illumination. leveraging the sensor's increased functionality, we propose to use inverse finite element method (ifem), a numerical method to reconstruct the contact force distribution based on marker displacements. the sensor is able to provide force distribution of contact with high spatial density. experiments and comparison with ground truth show that the reconstructed force distribution is physically reasonable with good accuracy.A sequence of Kendama manipulations with corresponding displacement field (yellow) and force field (red). Video can be found on Youtube: https://youtu.be/hWw9A0ZBZuU.

In this work, we develop a new surface reconstruction pipeline that combines monocular camera images and LiDAR measurements from a moving sensor rig to reconstruct dense 3D mesh models of indoor scenes. For surface reconstruction, the 3D LiDAR and camera are widely deployed for gathering geometric information from environments. Current state-of-the-art multi-view stereo or LiDAR-only reconstruction methods cannot reconstruct indoor environments accurately due to shortcomings of each sensor type. In our approach, LiDAR measurements are integrated into a multi-view stereo pipeline for point cloud densification and tetrahedralization. In addition to that, a graph cut algorithm is utilized to generate a watertight surface mesh. Because our proposed method leverages the complementary nature of these two sensors, the accuracy and completeness of the output model are improved. The experimental results on real world data show that our method significantly outperforms both the state-of-the-art camera-only and LiDAR-only reconstruction methods in accuracy and completeness.

Dense 3D visual mapping estimates as many as possible pixel depths, for each image. This results in very dense point clouds that often contain redundant and noisy information, especially for surfaces that are roughly planar, for instance, the ground or the walls in the scene. In this paper we leverage on semantic image segmentation to discriminate which regions of the scene require simplification and which should be kept at high level of details. We propose four different point cloud simplification methods which decimate the perceived point cloud by relying on class-specific local and global statistics still maintaining more points in the proximity of class boundaries to preserve the infra-class edges and discontinuities. 3D dense model is obtained by fusing the point clouds in a 3D Delaunay Triangulation to deal with variable point cloud density. In the experimental evaluation we have shown that, by leveraging on semantics, it is possible to simplify the model and diminish the noise affecting the point clouds.

In this paper, we propose a method for training control policies for human-robot interactions such as handshakes or hand claps via Deep Reinforcement Learning. The policy controls a humanoid Shadow Dexterous Hand, attached to a robot arm. We propose a parameterizable multi-objective reward function that allows learning of a variety of interactions without changing the reward structure. The parameters of the reward function are estimated directly from motion capture data of human-human interactions in order to produce policies that are perceived as being natural and human-like by observers. We evaluate our method on three significantly different hand interactions: handshake, hand clap and finger touch. We provide detailed analysis of the proposed reward function and the resulting policies and conduct a large-scale user study, indicating that our policy produces natural looking motions.

In this paper, a novel magnetic manipulation system using mobile coils for remote actuation of magnetic untethered devices in an enlarged workspace is proposed and studied. A parallel mechanism is implemented to actuate the mobile coils. A proof-of-concept prototype is designed and constructed, namely the DeltaMag, which includes three electromagnetic coils for generating magnetic fields and three motors for actuation of the coils. It has good space utilization: ratio between the diameter of the workspace and the diameter of the whole prototype reaches 0.7.A calibrated mathematical model is developed for the field distribution of a single coil, which has an average error of 8.75%. Then, we introduce a calculation method for the 3D magnetic field at any working position for the configuration of multiple parallel mobile coils. Moreover, an embedded system is established for actuating the parallel mechanism, whose pose is fed back via serial communication for magnetic field computation. A vision based approach is developed for closed-loop control of the parallel mechanism. Furthermore, experiments demonstrate the capabilities of the DeltaMag for manipulation of a magnetic catheter mock-up and a magnetic capsule mock-up in a workspace with a diameter more than 200 mm.

We present a novel method to control the relative positions of the members of a robotic team. The application scenario we consider is the cooperative manipulation of a deformable object in 2D space. A typical goal in this kind of scenario is to minimize the deformation of the object with respect to a desired state. Our contribution, then, is to use a global measure of deformation directly in the feedback loop. In particular, the robot motions are based on the descent along the gradient of a metric that expresses the difference between the team's current configuration and its desired shape. Crucially, the resulting multirobot controller has a simple expression and is inexpensive to compute, and the approach lends itself to analysis of both the transient and asymptotic dynamics of the system. This analysis reveals a number of properties that are interesting for a manipulation task: fundamental geometric parameters of the team (size, orientation, centroid, and distances between robots) can be suitably steered or bounded. We describe different policies within the proposed deformation-based control framework that produce useful team behaviors. We illustrate the methodology with computer simulations.

Detecting the intention of drivers is an essential task in self-driving, necessary to anticipate sudden events like lane changes and stops. Turn signals and emergency flashers communicate such intentions, providing seconds of potentially critical reaction time. In this paper, we propose to detect these signals in video sequences by using a deep neural network that reasons about both spatial and temporal information. Our experiments on more than a million frames show high per-frame accuracy in very challenging scenarios.

While the keypoint-based maps created by sparse monocular Simultaneous Localisation and Mapping (SLAM) systems are useful for camera tracking, dense 3D reconstructions may be desired for many robotic tasks. Solutions involving depth cameras are limited in range and to indoor spaces, and dense reconstruction systems based on minimising the photometric error between frames are typically poorly constrained and suffer from scale ambiguity. To address these issues, we propose a 3D reconstruction system that leverages the output of a Convolutional Neural Network (CNN) to produce fully dense depth maps for keyframes that include metric scale. Our system, DeepFusion, is capable of producing real-time dense reconstructions on a GPU. It fuses the output of a semi-dense multiview stereo algorithm with the depth and gradient predictions of a CNN in a probabilistic fashion, using learned uncertainties produced by the network. While the network only needs to be run once per keyframe, we are able to optimise for the depth map with each new frame so as to constantly make use of new geometric constraints. Based on its performance on synthetic and real world datasets, we demonstrate that DeepFusion is capable of performing at least as well as other comparable systems.

n-shot learning, i.e., learning a classifier from only few or even one training samples per class, is the ultimate goal in minimizing the cost of sample acquisition. This is esp. important for active sensing tasks like tactile material classification. Achieving high classification accuracy from only few samples is typically possible only when pre-knowledge is used. In n-shot transfer learning, knowledge from pre-training on a large knowledge set with many classes and samples per class has to be transferred to support the training for a given task set with only few samples per new class. In this paper, we show for the first time that deep end-to-end transfer learning is feasible for tactile material classification. Based on the previously presented (TactNet-II) [1], a deep convolutional neural network (CNN) which reaches superhuman tactile classification performance, we adapt state-of-the art deep transfer learning methods. We evaluate the resulting deep n-shot learning methods with a publicly available tactile material data set with 36 materials [1] in a 6-way n-shot learning task with 30 materials in the knowledge set. In 1-shot learning, our deep transfer learning method reaches 75.5% classification accuracy and in 10-shot more than 90%, outperforming classification without knowledge transfer by more than 40%. This results in an up to 15 time reduction in the number of samples needed to reach a desired accuracy level. We also provide insights of the inner workings of the derived deep transfer learning methods.

Estimation of tactile properties from vision, such as slipperiness or roughness, is important to effectively interact with the environment. These tactile properties help us decide which actions we should choose and how to perform them. E.g., we can drive slower if we see that we have bad traction or grasp tighter if an item looks slippery. We believe that this ability also helps robots to enhance their understanding of the environment, and thus enables them to tailor their actions to the situation at hand. We therefore propose a model to estimate the degree of tactile properties from visual perception alone (e.g., the level of slipperiness or roughness). Our method extends a encoder-decoder network, in which the latent variables are visual and tactile features. In contrast to previous works, our method does not require manual labeling, but only RGB images and the corresponding tactile sensor data. All our data is collected with a webcam and uSkin tactile sensor mounted on the end-effector of a Sawyer robot, which strokes the surfaces of 25 different materials. We show that our model generalizes to materials not included in the training data by evaluating the feature space, indicating that it has learned to associate important tactile properties with images.

Mobile robots are required to navigate freely in a complex and crowded environment in order to provide services to humans. For this navigation ability, deep reinforcement learning (DRL)-based methods are gaining increasing attentions. However, existing DRL methods require a wide field of view (FOV), which imposes the usage of high-cost lidar devices. In this paper, we explore the possibility of replacing expensive lidar devices with affordable depth cameras which have a limited FOV. First, we analyze the effect of a limited field of view in the DRL agents. Second, we propose a LSTM agent with Local-Map Critic (LSTM-LMC), which is a novel DRL method to learn efficient navigation in a complex environment with a limited FOV. Lastly, we introduce the dynamics randomization technique to improve the robustness of the DRL agents in the real world. We found that our method with a limited FOV can outperform the methods having a wide FOV but limited memory. We provide the empirical evidence that our method learns to implicitly model the surrounding environment and dynamics of other agents. We also show that a robot with a single depth camera can navigate through a complex real-world environment using our method.

While learning visuomotor skills in an end-to-end manner is appealing, deep neural networks are often uninterpretable and fail in surprising ways. For robotics tasks, such as autonomous driving, models that explicitly represent objects may be more robust to new scenes and provide intuitive visualizations. We describe a taxonomy of “object-centric” models which leverage both object instances and end-to-end learning. In the Grand Theft Auto V simulator, we show that object-centric models outperform object-agnostic methods in scenes with other vehicles and pedestrians, even with an imperfect detector. We also demonstrate that our architectures perform well on real-world environments by evaluating on the Berkeley DeepDrive Video dataset, where an object-centric model outperforms object-agnostic models in the low-data regimes.