ICRA 2018 论文列表

The forward kinematics of the Gough-Stewart platform, and their simplified versions in which some leg endpoints coalesce, has been typically solved using variable elimination methods. In this paper, we cast doubts on whether this is the easiest way to solve the problem. We will see how the indirect approach in which the length of some extra virtual legs is first computed leads to important simplifications. In particular, we provide a procedure to solve 30 out of 34 possible topologies for a Gough-Stewart platform without variable elimination.

This paper presents a six-limb high-speed parallel robot for pick-and-place operations that is based on two Delta robots, where the two sub-platforms are connected by a gearbox. Unlike the current 6-axis Delta-type robot, all the actuators of the robot are mounted on a base platform, allowing to reduce the inertia for high dynamic performance. Besides the 3-axis translations, the three rotations of the robot end-effector are realized by the differential motions of the two sub-platforms in three directions for large orientational workspace, but without significantly increased structural complexity of gearbox, compared to the existing one. The kinematic problems are studied to reveal that the workspace volume of the robot is similar to the commercial Delta-type robots. The simplified dynamic model is established and the simulation results show that the robot can reach up to a 20G acceleration subject to the commercial actuation combination.

When a space robot accidentally or voluntarily comes in contact with a target object, a workspace shift happens due to exchange of momentum between the objects. The problem of workspace adjustment is addressed herein. A novel controller is derived to simultaneously adjust the workspace and control the end-effector pose. The controller is based on a center-of-mass (CoM) regulation which fixes the workspace in the inertial space while leaving the base free to move, resulting in fuel efficiency. The control is validated on hardware using a robotic simulator composed of a seven degree-of-freedom (DOF) arm mounted on a 6DOF moving base.

Full Day Workshops & Tutorials

Most of the conventional approaches to mobile robot navigation avoid any kind of contact with the environment or with humans. As nowadays distance sensors typically have a limited - and often only two-dimensional - field of view, collisions with the environment or contacts with humans cannot be fully avoided in practical mobile robot applications. On the other hand, direct physical contact can be used for intuitive communication between a robot and humans. In this paper, we present a whole-body sensory concept based on a 6-DoF force-torque sensor to perceive physical interaction between the robot and humans. To distinguish between external contact and disturbance forces that result from the motion of the mobile platform or oscillations, we present a novel model-free filtering approach based on a neural network. In extensive experiments carried out with our robot Canny we demonstrate the effectiveness and advantages of the neural network approach, which clearly outperforms a classical model-based one.

Future planetary rovers will gain the ability to manipulate their environment in addition to the maneuverability of current systems. For dedicated contact interaction, Cartesian impedance control is a well-established approach from numerous terrestrial applications. In this paper we will present a whole-body Cartesian impedance controller for a planetary rover equipped with a robotic arm. In contrast to classical terrestrial whole-body controllers, the issue of proper wheel force distribution will be addressed within the control framework. A global optimization solves this redundancy in the over-actuation of the mobile base while additionally handling the kinematic redundancy in the serial kinematic sub-chain of the robot. The approach is experimentally validated on the DLR Lightweight Rover Unit. It can be used for versatile manipulation in rough terrain such as encountered in planetary exploration or terrestrial search-and-rescue scenarios.

This paper introduces and evaluates a new tensor field representation to express the geometric affordance of one object relative to another, a key competence for Cognitive and Autonomous robots. We expand the bisector surface representation to one that is weight-driven and that retains the provenance of surface points with directional vectors. We also incorporate the notion of affordance keypoints which allow for faster decisions at a point of query and with a compact and straightforward descriptor. Using a single interaction example, we are able to generalize to previously-unseen scenarios; both synthetic and also real scenes captured with RGB-D sensors. Evaluations also include crowdsourcing comparisons that confirm the validity of our affordance proposals, which agree on average 84 % of the time with human judgments, that is 20-40 % better than the baseline methods.

In this work, we propose a method for object recognition and pose estimation from depth images using convolutional neural networks. Previous methods addressing this problem rely on manifold learning to learn low dimensional viewpoint descriptors and employ them in a nearest neighbor search on an estimated descriptor space. In comparison we create an efficient multi-task learning framework combining manifold descriptor learning and pose regression. By combining the strengths of manifold learning using triplet loss and pose regression, we could either estimate the pose directly reducing the complexity compared to NN search, or use the learned descriptor for the NN descriptor matching. By in depth experimental evaluation of the novel loss function we observed that the view descriptors learned by the network are much more discriminative resulting in almost 30% increase regarding relative pose accuracy compared to related works. On the other hand, regarding directly regressed poses we obtained important improvement compared to simple pose regression. By leveraging the advantages of both manifold learning and regression tasks, we are able to improve the current state-of-the-art for object recognition and pose retrieval.

For achieving a successful mission in lunar exploration, not only traversability of exploration rover should be predicted but also operation environment should be considered under the limited power condition. Therefore, an optimal design of rover wheel for minimizing power consumption and maximizing tractive performance is required by conducting conceptual design stage. This paper describes settlement of requirements at system level and modeling of operation environment such as power acquisition and terrain characteristics in the lunar simulant. Using the wheel-terrain interaction model, a wheel design methodology was proposed to obtain an optimal wheel dimension, which meet the limited power condition along with maximal trafficability according to the each landing site. In addition, the results from the above approach have been validated with the experimental results using a single wheel test bed on the lunar simulant.

Scene models allow robots to reason about what is in the scene, what else should be in it, and what should not be in it. In this paper, we propose a hybrid Boltzmann Machine (BM) for scene modeling where relations between objects are integrated. To be able to do that, we extend BM to include tri-way edges between visible (object) nodes and make the network to share the relations across different objects. We evaluate our method against several baseline models (Deep Boltzmann Machines, and Restricted Boltzmann Machines) on a scene classification dataset, and show that it performs better in several scene reasoning tasks.

A challenge of pan/tilt/zoom (PTZ) camera networks for efficient and flexible visual monitoring is automated active network reconfiguration in response to environmental stimuli. In this paper, given an event/activity distribution over a convex environment, we propose a new provably correct reactive coverage control algorithm for PTZ camera networks that continuously (re) configures camera orientations and zoom levels (i.e., angles of view) in order to locally maximize their total coverage quality. Our construction is based on careful modeling of visual sensing quality that is consistent with the physical nature of cameras, and we introduce a new notion of conic Voronoi diagrams, based on our sensing quality measures, to solve the camera network allocation problem: that is, to determine where each camera should focus in its field of view given all the other cameras' configurations. Accordingly, we design simple greedy gradient algorithms for both continuous-and discrete-time first-order PTZ camera dynamics that asymptotically converge a locally optimal coverage configuration. Finally, we provide numerical and experimental evidence demonstrating the effectiveness of the proposed coverage algorithms.

There are a wide range of features that tactile contact provides, each with different aspects of information that can be used for object grasping, manipulation, and perception. In this paper inference of some key tactile features, tip displacement, contact location, shear direction and magnitude, is demonstrated by introducing a novel method of transducing a third dimension to the sensor data via Voronoi tessellation. The inferred features are displayed throughout the work in a new visualisation mode derived from the Voronoi tessellation; these visualisations create easier interpretation of data from an optical tactile sensor that measures local shear from displacement of internal pins (the TacTip). The output values of tip displacement and shear magnitude are calibrated to appropriate mechanical units and validate the direction of shear inferred from the sensor. We show that these methods can infer the direction of shear to ~2.3° without the need for training a classifier or regressor. The approach demonstrated here will increase the versatility and generality of the sensors and thus allow sensor to be used in more unstructured and unknown environments, as well as improve the use of these tactile sensors in more complex systems such as robot hands.

This paper proposes a new method, that we call VisualBackProp, for visualizing which sets of pixels of the input image contribute most to the predictions made by the convolutional neural network (CNN). The method heavily hinges on exploring the intuition that the feature maps contain less and less irrelevant information to the prediction decision when moving deeper into the network. The technique we propose is dedicated for CNN-based systems for steering self-driving cars and is therefore required to run in real-time. This makes the proposed visualization method a valuable debugging tool which can be easily used during both training and inference. We justify our approach with theoretical arguments and confirm that the proposed method identifies sets of input pixels, rather than individual pixels, that collaboratively contribute to the prediction. We utilize the proposed visualization tool in the NVIDIA neural-network-based end-to-end learning system for autonomous driving, known as PilotNet. We demonstrate that VisualBackProp determines which elements in the road image most influence PilotNet's steering decision and indeed captures relevant objects on the road. The empirical evaluation furthermore shows the plausibility of the proposed approach on public road video data as well as in other applications and reveals that it compares favorably to the layer-wise relevance propagation approach, i.e. it obtains similar visualization results and achieves order of magnitude speed-ups.

The development of fast, agile micro Unmanned Aerial Vehicles (UAVs) has been limited by (i) on-board computing hardware restrictions, (ii) the lack of sophisticated vision-based perception and vision-in-the-loop control algorithms, and (iii) the absence of development environments where such systems and algorithms can be rapidly and easily designed, implemented, and validated. Here, we first present a new micro UAV platform that integrates high-rate cameras, inertial sensors, and an NVIDIA Jetson Tegra X1 system-on-chip compute module that boasts 256 GPU cores. The UAV mechanics and electronics were designed and built in house, and are described in detail. Second, we present a novel “virtual reality” development environment, in which photorealistically-rendered synthetic on-board camera images are generated in real time while the UAV is in flight. This development environment allows us to rapidly prototype computing and sensing hardware as well as perception and control algorithms, using real physics, real interoceptive sensor data (e.g., from the on-board inertial measurement unit), and synthetic exteroceptive sensor data (e.g., from synthetic cameras). Third, we demonstrate repeated agile maneuvering with closed-loop vision-based perception and control algorithms, which we have developed using this environment.

This paper presents a novel strategy for autonomous visual saliency-aware receding horizon exploration of unknown environments using aerial robots. Through a model of visual attention, incrementally built maps are annotated regarding the visual importance and saliency of different objects and entities in the environment. Provided this information, a path planner that simultaneously optimizes for exploration of unknown space, and also directs the robot's attention to focus on the most salient objects, is developed. Following a two-step optimization paradigm, the algorithm first samples a random tree and identifies the branch maximizing for new volume to be explored. The first viewpoint of this path is then provided as a reference to the second planning step. Within that, a new tree is spanned, admissible branches arriving at the reference viewpoint while respecting a time budget dependent on the robot endurance and its environment exploration rate are found and evaluated in terms of reobserving salient regions at sufficient resolution. The best branch is then selected and executed by the robot, and the whole process is iteratively repeated. The proposed method is evaluated regarding its ability to provide increased attention toward salient objects, is verified to run onboard a small aerial robot, and is demonstrated in a set of challenging experimental studies.

In this paper we present minimal solutions for two-view relative motion estimation based on a homography formulation. By assuming a known vertical direction (e.g. from an IMU) and assuming a dominant ground plane we demonstrate that rotation and translation estimation can be decoupled. This result allows us to reduce the number of point matches needed to compute a motion hypothesis. We then derive different algorithms based on this decoupling that allow an efficient estimation. We also demonstrate how these algorithms can be used efficiently to compute an optimal inlier set using exhaustive search or histogram voting instead of a traditional RANSAC step. Our methods are evaluated on synthetic data and on the KITTI data set, demonstrating that our methods are well suited for visual odometry in road driving scenarios.

This study proposes a simple yet surprisingly effective feature matching approach, termed as guided locality preserving matching (GLPM), for visual homing of panoramic images. The key idea of our approach is merely to preserve the neighborhood structures of potential true matches between two panoramic images. We formulate it into a mathematical model, and derive a simple closed-form solution with linearithmic time and linear space complexities. This enables our method to accomplish the mismatch removal from hundreds of putative correspondences in only a few milliseconds. To handle extremely large proportions of outliers, we further design a guided matching strategy based on the proposed method, using the matching result on a small putative set with a high inlier ratio to guide the matching on a large putative set. This strategy can also significantly boost true matches without sacrifice in accuracy. To apply our GLPM to the visual homing problem, we develop a method for dense motion flow estimation from sparse feature matches based on Tikhonov regularization. Moreover, the focus-of-contraction/focus-of-expansion is derived to determine homing directions. The effectiveness of our method is demonstrated on a panoramic database in both feature matching and visual homing.

The grasping control of an aerial manipulator in practical environments is challenging due to its complex kinematics/dynamics and motion constraints. This paper introduces a lightweight aerial manipulator, which is combined with an X8 coaxial octocopter and a 4-DoF manipulator. To address the grasping control problem, we develop an efficient scheme containing trajectory generation, visual trajectory tracking, and kinematic compensation. The NSGA-II method is utilized to implement the multiobjective optimization for trajectory planning. Motion constraints and collision avoidance are also considered in the optimization. A kinematic compensation-based visual trajectory tracking is introduced to address the coupled nature between manipulator and VAV body. No dynamic parameter calibration is needed. Finally, several experiments are performed to verify the stability and feasibility of the proposed approach.

This paper focuses on visual tracking of a robotic manipulator during manipulation. In this situation, tracking is prone to failure when visual distractions are created by the object being manipulated and the clutter in the environment. Current state-of-the-art approaches, which typically rely on model-fitting using Iterative Closest Point (ICP), fail in the presence of distracting data points and are unable to recover. Meanwhile, discriminative methods which are trained only to distinguish parts of the tracked object can also fail in these scenarios as data points from the occlusions are incorrectly classified as being from the manipulator. We instead propose to use the per-pixel data-to-model associations provided from a random forest to avoid local minima during model fitting. By training the random forest with artificial occlusions we can achieve increased robustness to occlusion and clutter present in the scene. We do this without specific knowledge about the type or location of the manipulated object. Our approach is demonstrated by using dense depth data from an RGB-D camera to track a robotic manipulator during manipulation and in presence of occlusions.