ICRA 2015 论文列表

Natural muscles exhibit high power-to-weight ratios, inherent compliance and damping, fast actuation and high dynamic ranges. Unfortunately, traditional robotic actuators have been unable to attain similar properties, especially in a slender muscle-like form factor. Recently, super-coiled polymer (SCP) actuators have rejuvenated the promise of an artificial muscle. Constructed from commercial nylon fishing line or sewing thread and twisted until coils form, these lightweight actuators have been shown to produce significant mechanical power when thermally cycled. In this paper, we develop a thermomechanical and thermoelectric model of SCP actuators, and examine their controllability. With off-the-shelf conductive sewing thread, we show the ability to produce controlled forces in under 30 ms, exceeding human muscle performance. Finally, we use SCP actuators in a robotic hand to demonstrate their applicability as a low-cost, high performance robotic muscle.

This paper reports on a method to perform robust visual relocalization between temporally separated sets of underwater images gathered by a robot. The place recognition and relocalization problem is more challenging in the underwater environment mainly due to three factors: 1) changes in illumination; 2) long-term changes in the visual appearance of features because of phenomena like biofouling on man-made structures and growth or movement in natural features; and 3) low density of visually salient features for image matching. To address these challenges, a patch-based feature matching approach is proposed, which uses image segmentation and local intensity contrast to locate salient patches and HOG description to make correspondences between patches. Compared to traditional point-based features that are sensitive to dramatic appearance changes underwater, patch-based features are able to encode higher level information such as shape or structure which tends to persist across years in underwater environments. The algorithm is evaluated on real data, from multiple years, collected by a Hovering Autonomous Underwater Vehicle for ship hull inspection. Results in relocalization performance across missions from different years are compared to other traditional methods.

Learning from demonstration (LfD) aims at robots learning skills from human-demonstrated tasks. Robots should be able to learn at all levels of abstraction. Unlike at the level of motor primitives, high-level LfD requires symbolic representations. It thus faces the classical problem of symbol grounding. Furthermore, it requires the robot to interpret human-demonstrated actions at a higher, conceptual abstraction level. We present a method, that enables a robot to recognize human-demonstrated pick-and-place task goals on an object-relational abstraction layer. The robot can reproduce the task goals in new situations using a symbolic planner. We show that in a robotic context conceptual spaces can serve as a mean for symbol grounding at an object-relational level as well as for the recognition of conceptual similarities in effects of human-demonstrated actions. The method is evaluated in experiments on a real robot.

In ambient intelligence, pressure sensing can be used for detecting and recognizing objects based on their load profile. This paper presents a pressure scanning technique that improves weight-based object recognition, by adding information about the surface of the object in contact with the floor. The new high-resolution pressure scanning technique employs sub-pixel shifting to assemble a series of low-resolution scans into an aggregated high-resolution scan. The proposed scanning device is composed of 4 load-sensing tiles, on which the scanned object slides in regular movements. The result is a regular grid image of the object's contact surface, containing the weight of each section of the grid, as well as the corresponding centers of mass. A formal proof-of-concept is provided, together with experimental results obtained both on a noiseless simulated platform, and on a noisy physical platform.

This paper presents a novel quadrotor UAV attitude control algorithm to realize complex acrobatic UAV maneuvers. A nonlinear dynamic model based on the exponential coordinates parametrization of rotation is proposed. By analysing the model using Lie Group and Lie Algebra theory, cascaded linear PID controllers are designed. To further improve the controller performance, PID controllers are augmented with smith predictor and rotational trajectory planner. The experiments conducted on a real quadrotor show that our control algorithm surpasses most known quadrotor controllers.

We propose an hierarchical control framework for multiple cooperative quadrotor-manipulator systems, which allows us to endow the common grasped object with a user-specified desired behavior (e.g., trajectory tracking, compliant interaction, etc.). To achieve this, our control framework consists of the following hierarchical layers: 1) object desired behavior design; 2) optimal cooperative force distribution; and 3) individual quadrotor-manipulator control based on object stiffness model, which can also take into account different dynamics characteristics of the (slower/coarse) quadrotor-platform and the (faster/fine) manipulator. Simulations of object transport and compliant interaction with three quadrotor-manipulator systems are performed to illustrate the theory.

Similar to combustion engines comprising multiple cylinders engaged with a crankshaft, multiple piezoelectric stack actuators (PSA) engaged with a common output rod can produce smooth, long stroke motion with desired properties. In particular, when equally spaced multiple units are arranged to push sinusoidal gear teeth on the output rod, the system exhibits unique collective behaviors thanks to “harmonic” effects of the multiple units. For example, although the force-displacement characteristics of individual units are highly nonlinear, the undesirable nonlinearity, including singularity, may be eliminated. Here we present harmonic analysis and control of a class of actuators consisting of multiple driving units engaged with a sinusoidal transmission, termed a harmonic poly-actuator. Through theoretical analysis we obtain 1) conditions on the unit arrangement to eliminate their nonlinearity from the output force, 2) control algorithms for coordinating the multiple units to generate a commanded force with desired force-displacement characteristics, and 3) a method for compensating for output force ripples due to possible misalignment and heterogeneity of individual units. The control algorithms are implemented on a prototype harmonic poly-actuator with six units of PSAs. Experiments demonstrate the unique features of the poly-actuator exploiting the harmonic properties of the system.

Constrained Markov Decision Processes offer a principled way to tackle sequential decision problems with multiple objectives. Although they could be very valuable in numerous robotic applications, to date their use has been quite limited. One of the reasons is that their solution requires to solve constrained linear programs with a large number of variables and this is computationally demanding, especially when considering dynamic environments. In this paper we propose a hierarchical approach to solve large CMDPs. States are clustered into macro states and relevant parameters like transition probabilities and costs are extracted with a Monte Carlo approach. Macro states are created with the objective of grouping together states with similar costs while preserving feasibility. We illustrate the value of our findings in a path planning scenario where the robot moves through an environment characterized by different risk levels. Our approach largely outperforms the non-hierarchical method and we also show how it prevails over methods based on fixed partitioning strategies.

We propose a technique for joint calibration of a wide-angle rolling shutter camera (e.g. a GoPro) and an externally mounted gyroscope. The calibrated parameters are time scaling and offset, relative pose between gyroscope and camera, and gyroscope bias. The parameters are found using non-linear least squares minimisation using the symmetric transfer error as cost function.

We propose a language-driven navigation approach for commanding mobile robots in outdoor environments. We consider unknown environments that contain previously unseen objects. The proposed approach aims at making interactions in human-robot teams natural. Robots receive from human teammates commands in natural language, such as “Navigate around the building to the car left of the fire hydrant and near the tree”. A robot needs first to classify its surrounding objects into categories, using images obtained from its sensors. The result of this classification is a map of the environment, where each object is given a list of semantic labels, such as “tree” and “car”, with varying degrees of confidence. Then, the robot needs to ground the nouns in the command. Grounding, the main focus of this paper, is mapping each noun in the command into a physical object in the environment. We use a probabilistic model for interpreting the spatial relations, such as “left of” and “near”. The model is learned from examples provided by humans. For each noun in the command, a distribution on the objects in the environment is computed by combining spatial constraints with a prior given as the semantic classifier's confidence values. The robot needs also to ground the navigation mode specified in the command, such as “navigate quickly” and “navigate covertly”, as a cost map. The cost map is also learned from examples, using Inverse Optimal Control (IOC). The cost map and the grounded goal are used to generate a path for the robot. This approach is evaluated on a robot in a real-world environment. Our experiments clearly show that the proposed approach is efficient for commanding outdoor robots.

Efficient estimation of obstacles is a crucial component in autonomous vehicle navigation. Fusing stereo vision and laser scan data is a widely used approach for obstacle detection and navigation. Stereo vision can detect obstacles and the ground. Planar laser range finders can detect obstacles, but false positives occur when the scans collide with the ground. This paper proposes an algorithm for cooperative fusion of stereo vision and laser data. The algorithm checks the consensus of the two signals before filtering out the false positive laser scans from ground hits. Experimental results for tests in real driving situations in an industrial environment are presented. The results confirm the effectiveness of the proposed algorithm for robust and safe navigation.

Nearly all robotic grippers have one trait in common: they grasp objects with normal forces, either directly, or indirectly through friction. This method of grasping is effective for objects small enough for a given gripper to partially encompass. However, to grasp larger objects, significant grip forces and a high coefficient of friction are required. We present a new grasping method for convex objects, using almost exclusively shear forces. We achieve shear grasping with a gripper that utilizes thin film gecko-inspired fibrillar adhesives that conform to the curvature of the object. We present a verified model for grasping a range of curvatures and results that demonstrate the thin film fibrillar adhesives' increased contact area on textured surfaces when loaded in shear. Finally, the gripper is implemented on a robotic arm and grasps a variety of convex objects (at rest and ballistic).

In many scenarios, robots encounter rotationally symmetric objects for which no known 3D model exists. To be able to grasp such objects using existing grasp point computation schemes, an estimate of their 3D-pose and shape is necessary. In this paper, we address the problem of recovering 3D-pose and shape of an unknown surface of revolution from two perspective views of known relative orientation. We propose a new algorithm for simultaneous estimation of pose and shape without making use of any cross-sections or bi-tangent points needed by other approaches. Our algorithm builds upon existing single-view SOR reconstruction approaches and couples the pose estimation and reconstruction process. Pose is optimized to minimize discrepancies between reconstructed shapes from two views. Our method works even in the presence of only one of the two apparent contours of a surface of revolution. We test our algorithm by comparing to ground-truth poses and shapes as well as performing grasping experiments. We introduce a new dataset of rotationally symmetric objects in a variety of poses and backgrounds and with a measured ground-truth pose and shape.

This study presents a robot-hand-arm system with high robustness and responsiveness by using a “Net-Structure Proximity Sensor.” The sensor, which we have developed and specially designed for a robot hand, directly detects an object being to be grasped and outputs analog voltage signals according to the position/posture error between the robot hand and the object. It has been confirmed that the robot hand is able to quickly adjust to and grasp an unknown object by applying a feed-back control method based on the sensor signals. This paper focuses on the integration of the proximity-based feedback control to a commonly-used vision-based control. These sensors work in complementary manner: a vision sensor is available for planning an approaching path of a robot hand by detecting large area, and a Net-Structure Proximity Sensor enables the robot hand to adjust the approaching error before grasping and to improve the certainty of the grasping. Two objective velocities are derived independently by the sensors. By adding the velocities with considering the reliability of the sensor information, the robot hand becomes to be able to perform approaching and adjustment to the target object simultaneously. Experimental results showed that the robot hand grasped a moving object with high success rate even in conditions where it was difficult to predict the trajectory of the object accurately.

The graph-based SLAM (Simultaneous Localization and Mapping) method uses a graph to represent and solve the SLAM problem. The SLAM allows building a map of an unknown environment and simultaneously localizing the robot on this map. This paper presents a temporal analysis of the 3D graph-based SLAM method. We also propose an efficient implementation, on an OMAP embedded architecture, which is a widely used open multimedia applications platform. We provide an optimized data structure and an efficient memory access management to solve the nonlinear least squares problem related to the algorithm. The algorithm takes advantage of the Schur complement to reduce the execution time. We will present an optimized implementation of this task. We also take advantage of the multi-core architecture to parallelize the algorithm. To evaluate our implementation, we will compare the computational performances to the well known framework g2o. This work aims to demonstrate how optimizing data structure and multi-threading can decrease significantly the execution time of the graph-based SLAM on a low-cost architecture dedicated to embedded applications.

We propose a methodology for learning and using a multiple-goal probabilistic motion model within a particle filter-based target tracking on video streams. In a set of training video sequences, we first extract the locations (coined as “goals”) where the pedestrians either leave the scene or often change directions. Then, we learn one motion prior model per detected goal. Each of these models is learned statistically based on the local motion observed by the camera during the training phase. Given that the initial, empirical distribution may be incomplete or noisy, we regularize it in a second phase. These priors are then used in an Interactive Multiple Model (IMM) scheme for target tracking and goal estimation. We demonstrate the relevance of this methodology with tracking experiments and comparisons done on standard datasets.

In this paper we present an algorithm to determine the location of contact points to obtain force closure grasps on tree dimensional objects. The shape of the object is assumed to be given by a triangle mesh - a format widely used in CAD software. Our algorithm can handle an arbitrary number of contact points and does nor require any prior information about their initial locations. Through an iterative process, contact point locations are updated aiming at improving a commonly used grasp quality metric. The process is global in the sense that during the process the whole surface of the object can be explored, and contact point locations can cross sharp edges that usually represent a problem for optimization algorithms relying on smooth surface representations. Extensive simulation results illustrate the performance of the proposed method, outlining strengths and directions for further research.

In this paper, we consider the problem of connectivity maintenance in multi-robot systems with unicycle kinematics. While previous work has approached this problem through local control techniques, we propose a solution which achieves global connectivity maintenance under nonholonomic constraints. In addition, our formulation only requires intermittent estimation of algebraic connectivity, and accommodates discontinuous spatial interactions among robots. Specifically, we extend a decision-based link maintenance framework to unicycle kinematics and discontinuous potential-based interaction, by exploiting techniques from nonsmooth analysis. Then, we couple this extension with an existing connectivity estimation technique which yields an estimate with tunable precision in finite time, achieving our result. To illustrate the correctness of our methods, we provide a brief simulation result that closes the paper.

Gesture recognition is used for many practical applications such as human-robot interaction, medical rehabilitation and sign language. In this paper, we apply a hybrid generative-discriminative approach by using the Fisher Vector to improve the recognition performance. The strategy is to merge the generative approach of Hidden Markov Model dealing with spatio-temporal motion data with the discriminative approach of Support Vector Machine focusing on the classification task. The motion segments are encoded into HMMs, and each segment is converted to FV, whose elements can be obtained as the derivative of the probability of the segment being generated by the HMMs with respect to their parameters. SVM is subsequently trained by the FVs. An input gesture can be classified to corresponding gesture category by SVM. In the experiments, we test our approach by comparing three HMM chain models and four categorization methods on dataset provided by the ChaLearn Looking at People Challenge 2014 (LAP 2014). The results show that similar gesture patterns are clustered closely in several categories. Our approach based left-to-right HMMs outperforms other gesture recognition methods. More specifically, the hybrid generative-discriminative approach overcomes the standard HMM approach and the generative kernel approach overcomes the generative embedding approach. For these results, our approach is effective to improve the recognition performance.