IROS 2017 论文列表

The determination of a flexible needle shape during the insertion is an important issue in minimally invasive surgery techniques. This is especially critical when considering conventional surgery procedures where a surgeon usually determines how to proceed needle insertion further based on the current needle trajectory inserted into tissue during biopsy and removal of malignant tissues in the body. In this paper, we propose a new method to determine the shape of a needle that is being inserted, together with the curvature measurement data obtained by fiber Bragg gratings (FBG) sensors inside the needle. This approach can be particularly advantageous to the situations where visual guidance (such as by ultrasound probes) is not easily applicable. The description of a needle shape is based on the elastic rod theory and Lie-group-theoretic approach. We also present the comparison between two different calibration methods, which have an impact on the quality of the results of the proposed method. In order to verify the proposed method, needle trajectories by the model are compared with experimental data obtained by image analysis, which in turn emphasizes the capability of the proposed method.

This work provides an architecture to enable robotic grasp planning via shape completion. Shape completion is accomplished through the use of a 3D convolutional neural network (CNN). The network is trained on our own new open source dataset of over 440,000 3D exemplars captured from varying viewpoints. At runtime, a 2.5D pointcloud captured from a single point of view is fed into the CNN, which fills in the occluded regions of the scene, allowing grasps to be planned and executed on the completed object. Runtime shape completion is very rapid because most of the computational costs of shape completion are borne during offline training. We explore how the quality of completions vary based on several factors. These include whether or not the object being completed existed in the training data and how many object models were used to train the network. We also look at the ability of the network to generalize to novel objects allowing the system to complete previously unseen objects at runtime. Finally, experimentation is done both in simulation and on actual robotic hardware to explore the relationship between completion quality and the utility of the completed mesh model for grasping.

Tactile object shape identification is important for robotic hands to perform dexterous manipulation. Most of the proposed approaches concentrate on specific object recognition. This limits their application into more realistic environments where a larger amount of objects are present. We present a method that performs object shape identification independently on the size and location of the object within the hand. This method allows sequential learning of new shapes. The method combines proprioceptive signatures and contact normal information to build a descriptor that reduces the impact of the size and pose of the object on recognition rate. Sequential training is performed with a collection of Neural Forests (NF). This allows adding new objects to the training set so that training the model from scratch is avoided. Extensive experiments reveal that the combination of multiple modalities (e.g. contact normals, proprioceptive information) is beneficial to the system accuracy, and that results for sequential learning are on par with its batch counterpart. This makes sequential training advantageous because it is less time consuming. Results showed that both techniques depict similar results and perform with at least 83% in a 7-shapes case scenario in a simulated environment. Experiments are made with a real shadow hand using a NF trained on simulated data.

Sampling-based algorithms offer computationally efficient, practical solutions to the path finding problem in high-dimensional complex configuration spaces by approximately capturing the connectivity of the underlying space through a (dense) collection of sample configurations joined by simple local planners. In this paper, we address a long-standing bottleneck associated with the difficulty of finding paths through narrow passages. Whereas most prior work considers the narrow passage problem as a sampling issue (and the literature abounds with heuristic sampling strategies) very little attention has been paid to the design of new effective local planners. Here, we propose a novel sensory steering algorithm for sampling-based motion planning that can “feel” a configuration space locally and significantly improve the path planning performance near difficult regions such as narrow passages. We provide computational evidence for the effectiveness of the proposed local planner through a variety of simulations which suggest that our proposed sensory steering algorithm outperforms the standard straight-line planner by significantly increasing the connectivity of random motion planning graphs.

We present a novel end-to-end learning framework to enable ground vehicles to autonomously navigate unknown environments by fusing raw pixels from cameras and depth measurements from a LiDAR. A deep neural network architecture is introduced to effectively perform modality fusion and reliably predict steering commands even in the presence of sensor failures. The proposed network is trained on our own dataset, from LiDAR and a camera mounted on a UGV taken in an indoor corridor environment. Comprehensive experimental evaluation to demonstrate the robustness of our network architecture is performed to show that the proposed deep learning neural network is able to autonomously navigate in the corridor environment. Furthermore, we demonstrate that the fusion of the camera and LiDAR modalities provides further benefits beyond robustness to sensor failures. Specifically, the multimodal fused system shows a potential to navigate around static and dynamic obstacles and to handle changes in environment geometry without being trained for these tasks.

Deep reinforcement learning is becoming increasingly popular for robot control algorithms, with the aim for a robot to self-learn useful feature representations from unstructured sensory input leading to the optimal actuation policy. In addition to sensors mounted on the robot, sensors might also be deployed in the environment, although these might need to be accessed via an unreliable wireless connection. In this paper, we demonstrate deep neural network architectures that are able to fuse information generated by multiple sensors and are robust to sensor failures at runtime. We evaluate our method on a search and pick task for a robot both in simulation and the real world.

Although there is an abundance of planar fiducial-marker systems proposed for augmented reality and computer-vision purposes, using them to estimate the pose accurately in robotic applications where collected data are noisy remains a challenge. This is inherently a difficult problem because these fiducial marker systems work solely within the RGB image space and the resolution of cameras on robots is often constrained. As a result, small noise in the image would cause the tag's detection process to produce large pose estimation errors. This paper describes an algorithm that improves the pose estimation accuracy of square fiducial markers in difficult scenes by fusing information from RGB and depth sensors. The algorithm retains the high detection rate and low false positive rate characteristics of fiducial systems while making them much more robust to size, lighting and sensory noise for pose estimation. The improvements make the fiducial tags suitable for robotic tasks requiring high pose accuracy in the real world environment.

One of the main tasks in robotics is to precisely connect the robot's coordinate system with the world's coordinate system. For parallel mechanisms, this is a challenging task due to the direct kinematics problem, which is usually solved by using iterative algorithms to search for the mobile platform's position and orientation or by adding auxiliary sensors. In this paper, a sensor concept for solving the direct kinematics problem of the Stewart-Gough platform is presented where only three inertial measurement units are used with the aim to renounce length sensors as well as initial and, in some cases, external pose measurements.

Grasping and manipulation of biological tissue are crucial processes during minimally invasive surgery (MIS). To enable atraumatic and reliable grasping, it would be useful to detect slip of the grasped object. Because tissue is moist, conformable, and delicate, and because the sensor must work in a surgical environment, this application requires a departure from conventional slip sensing methods. We present a technology and method based on hot-wire anemometry to detect slip while grasping wet, conformable materials and discuss how this approach may be extended to graspers used in robot-assisted surgery (RAS). We present our design and the results from characterization tests as well as experimental results that demonstrate its ability to detect planar direction of slip of wet, compliant objects.

Small unmanned aircraft systems (UAS) must be able to detect and avoid conflicting traffic, an especially challenging task when the threat is another small UAS. Collision avoidance requires trajectory prediction and the performance of a collision avoidance system can be improved by extending the prediction horizon. We describe an algorithm that predicts the trajectory of a small, fixed-wing UAS using an estimate of its orientation. First, a computer vision algorithm locates specific feature points of the threat aircraft in an image. Next, the POSIT algorithm uses these feature points to estimate the pose (position and attitude) of the threat. A sequence of pose estimates is then used to predict the trajectory of the threat aircraft in order to avoid a collision. To assess the algorithm's performance, the predictions are compared with predictions based solely on position estimates for a variety of encounter scenarios. Simulation and experimental results indicate that trajectory prediction using orientation estimates provides quicker response to a change in the threat aircraft trajectory and better prediction and avoidance performance.

Precision farming robots offer a great potential for reducing the amount of agro-chemicals that is required in the fields through a targeted, per-plant intervention. To achieve this, robots must be able to reliably distinguish crops from weeds on different fields and across growth stages. In this paper, we tackle the problem of separating crops from weeds reliably while requiring only a minimal amount of training data through a semi-supervised approach. We exploit the fact that most crops are planted in rows with a similar spacing along the row, which in turn can be used to initialize a vision-based classifier requiring only minimal user efforts to adapt it to a new field. We implemented our approach using C++ and ROS and thoroughly tested it on real farm robots operating in different countries. The experiments presented in this paper show that with around 1 min of labeling time, we can achieve classification results with an accuracy of more than 95% in real sugar beet fields in Germany and Switzerland.

Place categorisation; i.e., learning to group perception data into categories based on appearance; typically uses supervised learning and either visual or 2D range data. This paper shows place categorisation from 3D data without any training phase. We show that, by leveraging the NDT histogram descriptor to compactly encode 3D point cloud appearance, in combination with standard clustering techniques, it is possible to classify public indoor data sets with accuracy comparable to, and sometimes better than, previous supervised training methods. We also demonstrate the effectiveness of this approach to outdoor data, with an added benefit of being able to hierarchically categorise places into sub-categories based on a user-selected threshold. This technique relieves users of providing relevant training data, and only requires them to adjust the sensitivity to the number of place categories, and provide a semantic label to each category after the process is completed.

This paper addresses the problem of generating coverage paths-that is, paths that pass within some sensor footprint of every point in an environment-for vehicles with Dubins motion constraints. We extend previous work that solves this coverage problem as a traveling salesman problem (TSP) by introducing a practical heuristic algorithm to reduce runtime while maintaining near-optimal path length. Furthermore, we show that generating an optimal coverage path is NP-hard by reducing from the Exact Cover problem, which provides justification for our algorithm's conversion of Dubins coverage instances to TSP instances. Extensive experiments demonstrate that the algorithm does indeed produce length paths comparable to optimal in significantly less time.

This paper specifies a surgical robot performing semi-autonomous electrosurgery for tumor resection and evaluates its accuracy using a visual servoing paradigm. We describe the design and integration of a novel, multi-degree of freedom electrosurgical tool for the smart tissue autonomous robot (STAR). Standardized line tests are executed to determine ideal cut parameters in three different types of porcine tissue. STAR is then programmed with the ideal cut setting for porcine tissue and compared against expert surgeons using open and laparoscopic techniques in a line cutting task. We conclude with a proof of concept demonstration using STAR to semi-autonomously resect pseudo-tumors in porcine tissue using visual servoing. When tasked to excise tumors with a consistent 4mm margin, STAR can semi-autonomously dissect tissue with an average margin of 3.67 mm and a standard deviation of 0.89mm.

Point cloud registration is a core problem of many robotic applications, including simultaneous localization and mapping. The Normal Distributions Transform (NDT) is a method that fits a number of Gaussian distributions to the data points, and then uses this transform as an approximation of the real data, registering a relatively small number of distributions as opposed to the full point cloud. This approach contributes to NDT's registration robustness and speed but leaves room for improvement in environments of limited structure. To address this limitation we propose a method for the introduction of semantic information extracted from the point clouds into the registration process. The paper presents a large scale experimental evaluation of the algorithm against NDT on two publicly available benchmark data sets. For the purpose of this test a measure of smoothness is used for the semantic partitioning of the point clouds. The results indicate that the proposed method improves the accuracy, robustness and speed of NDT registration, especially in unstructured environments, making NDT suitable for a wider range of applications.

This paper proposes a method for jointly estimating road layout and segmenting urban scenes semantically by applying a relative location prior. The proposed method is based on the conjecture that some relevant elements in urban environments tend to be located in a predictable manner. This belief can be modeled as a location prior to help a segmentation algorithm infer latent labels in images more accurately. In order to apply these structural characteristics, a set of special coordinates, referred to as road-normal coordinates, are defined which are perpendicular to the road. These coordinates are determined by estimating the most fittable road layout based on a marginal probability obtained from an existing segmentation algorithm. All possible segments in an image are projected into the road-normal coordinates with the aid of depth information from sensor measurements, and the pre-trained location prior is applied to each segment as an additional potential of a conditional random field (CRF) model. The proposed method is evaluated on the publicly available KITTI dataset including images and corresponding 3D point clouds.

Semantic 3D mapping can be used for many applications such as robot navigation and virtual interaction. In recent years, there has been great progress in semantic segmentation and geometric 3D mapping. However, it is still challenging to combine these two tasks for accurate and large-scale semantic mapping from images. In the paper, we propose an incremental and (near) real-time semantic mapping system. A 3D scrolling occupancy grid map is built to represent the world, which is memory and computationally efficient and bounded for large scale environments. We utilize the CNN segmentation as prior prediction and further optimize 3D grid labels through a novel CRF model. Superpixels are utilized to enforce smoothness and form robust PN high order potential. An efficient mean field inference is developed for the graph optimization. We evaluate our system on the KITTI dataset and improve the segmentation accuracy by 10% over existing systems.

This paper presents a self-supervised online learning approach for 3D object tracking that requires no pretraining of appearance. Our method focuses on selecting the most relevant parts of the RGBD input by continuously updating appearance classifiers in conjunction with the spatial occupancy of the target. Fine-grained regions selected via the learned bottom-up saliency, together with spatial cues of the 3D shape model, are used to identify and localize the target via shape registration. The subsequent 3-D pose estimate along with positive and negative labels from the registration are used for online learning appearance. The proposed method outperforms competing model-based tracking algorithms on public datasets as well as on a new motion scene dataset that we have collected.

This paper investigates self-reconfiguration of a modular robotic system, which consists of a cluster of modular vehicles that can attach to each other by a connection mechanism. Thereby, they can form a desired morphology to meet task specific requirements. Reconfiguration can be needed due to limitations from dimensions of passable corridors for an underwater maintenance task, for supplemental instrumentation that is available on a particular robot, or as remedial action if one robot in a cluster suffers from malfunction. Being crucial for autonomous underwater vehicles, energy consumed is employed as a heuristic. The paper shows how the Basic Theta* algorithm can be guided by an energy criterion to calculate a transition from start- to goal morphology. Individual robots are guided while minimizing the overall energy for propulsion and for balancing restoring forces and moments in morphologies. The properties of the proposed self-reconfiguration algorithm are evaluated through simulations and preliminary model tank experiments. The energy based heuristic for reconfiguration is compared to a traditional solution that minimizes the Euclidean distance.