IROS 2015 Papers

This work presents a soft hand capable of robustly grasping and identifying objects based on internal state measurements. A highly compliant hand allows for intrinsic robustness to grasping uncertainty, but the specific configuration of the hand and object is not known, leaving undetermined if a grasp was successful in picking up the right object. A soft finger was adapted and combined to form a three finger gripper that can easily be attached to existing robots, for example, to the wrist of the Baxter robot. Resistive bend sensors were added within each finger to provide a configuration estimate sufficient for distinguishing between a set of objects. With one data point from each finger, the object grasped by the gripper can be identified. A clustering algorithm to find the correspondence for each grasped object is presented for both enveloping grasps and pinch grasps. This hand is a first step towards robust proprioceptive soft grasping.

We propose a novel method for performing fine-grained recognition of human hand grasp types using a single monocular image to allow computational systems to better understand human hand use. In particular, we focus on recognizing challenging grasp categories which differ only by subtle variations in finger configurations. While much of the prior work on understanding human hand grasps has been based on manual detection of grasps in video, this is the first work to automate the analysis process for fine-grained grasp classification. Instead of attempting to utilize a parametric model of the hand, we propose a hand parsing framework which leverages a data-driven learning to generate a pixel-wise segmentation of a hand into finger and palm regions. The proposed approach makes use of appearance-based cues such as finger texture and hand shape to accurately determine hand parts. We then build on the hand parsing result to compute high-level grasp features to learn a supervised fine-grained grasp classifier. To validate our approach, we introduce a grasp dataset recorded with a wearable camera, where the hand and its parts have been manually segmented with pixel-wise accuracy. Our results show that our proposed automatic hand parsing technique can improve grasp classification accuracy by over 30 percentage points over a state-of-the-art grasp recognition technique.

Recently, there are an increasing number of developing wearable devices that intent to display various contents. As can be seen in sign languages and hand signals, hand gestures can express rich information with small motions, therefore, they are useful input for those devices. Especially, these gestures are less obtrusive actions than other actions such as speaking. It means that hand-gesture-based input interfaces are very suitable for public use, where a user can't use desktop devices such as a mouse or a keyboard. Although there are many studies on recognition of hand gestures, existing hand shape recognition methods have several problems to be applied for home automation. A wearable wrist contour sensor device has been proposed in previous studies. However discussions about how to design the protocols of hand gesture interface are insufficient. In this paper, we develop real time hand gesture interfaces of content browse for wearable or remote displays. Usability test using the developed interfaces reveals that pronation is to be assigned as a variable configurator and hand shape is to be assigned as an operation switcher.

Human Odometry (HO) is the process of providing a person with a continuous estimate of their location, on the basis of information acquired solely by sensors carried around by the person themselves. In an effort towards the development of effective and robust HO systems, we present the Human Odometry Outdoor Dataset (HOOD), a public collection of labelled accelerometer and gyroscope data recordings. We compare four sensor placements (foot, waist, wrist, chest) to identify the most suitable placement for different types of motions (ranging from walking to slithering), occurring in highly diverse real environments (such as flat grass fields, staircases and rough terrains).

One long term goal of robotic research is the development of robot systems, which have approximately the same cognitive, communicational, and manipulative abilities like humans and which can be operated by non-experts. The usage by non-experts sets additional requirements to the abilities of a robot system. For instance, the system should be able to understand natural language as one of the primary communication forms. This requires knowledge about actions, which shall be executed, or about objects, which shall be manipulated. In this work, we focus on grounding knowledge for actions within the domain of object manipulation. Our approach is based on the concept of verbalized physical effects and manipulation primitives. Therefore, we introduce the concept of verbalized physical effects, describe the relations between verbal instructions, verbalized physical effects and sensor based robot motions, and evaluate our approach using a robot manipulator.

This paper presents an account of the experimentations with a ground-based facility emulating interception scenarios with non-cooperative space debris using a robotic manipulator. A sampling-based motion-planning algorithm is used to autonomously solve the dynamic interception problem without collisions and under velocity, acceleration and jerk constraints. Invariant Kalman filter methodologies are employed to estimate and predict the motion of a neutrally-buoyant airship emulating a free-floating target. Discussions are presented on the mitigation of practical issues of system integration and the operation of a sensitive, but cost-effective, test-bed for aerial neutral-buoyancy experiments. An account of ground-based experiments is presented as well as results showing good success rates on the tested scenarios and methods that constitute the first reported set of experiments on the interception phase of active space debris removal.

This paper proposes a novel technique for segmenting the ground plane and at the same time estimating the occupancy probability of each point in a scene. Using a stereo camera rig, our system first calculates a disparity map and transforms it to a v-disparity map, which is then filtered and processed to generate a corresponding probability field. The probability field generated is then used for precise segmentation of ground planes as well as for the generation of occupancy grids. Unlike what is proposed in the prior art, our system requires minimal initialization and is independent of the stereo sensor characteristics as well as the parameters of the disparity algorithm. More importantly, our technique does not require any prior assumption about the terrain visual characteristics. Experimental results using sequences of images from two different data sets are presented to validate the proposed methods.

Quick motion and soft touch control are important for autonomous grasping. To perform a grasping action, a hand must adjust its fingertips to match the object shape. It is also necessary to reduce the fingertip velocity on contact with the object.We propose a method of fingertip velocity control for fast approach and slow contact using proximity sensors installed on fingertips. The proposed control reduces the fingertip velocity at contact, decreasing the change of impulse force and thus making it appropriate for grasping soft or fragile objects. The proposed control uses time-to-contact (TTC), which is converted from sensor output. It is known that many animals, including humans, use TTC for collision avoidance. TTC represents the remaining time until collision considering the rate change of a control variable. We carried out grasping tests on various common objects using TTC. Experimental results show that the proposed control realizes fingertip alignment perpendicular to unknown shapes and surface objects and lowers the velocity at contact.

We propose an evaluation method of grasp stability which takes into account the elastic deformation of fingertips from the viewpoint of energy. An evaluation value of grasp stability is derived as the minimum energy which causes slippage of a fingertip on its contact surface. To formulate the evaluation value, the elastic potential energy of fingertips and the gravitational potential energy of a grasped object are considered. It is ensured that fingertips do not slip on grasped object surfaces if the external energy applied to the object is less than the evaluation value. Since our evaluation value explicitly considers the deformation values of fingertips, grasp stability is evaluated by taking into consideration the contact forces generated by the deformation. The effectiveness of our method is verified through numerical examples.

Knowledge of the human grasp pose is crucial in common control schemes for human-robot object manipulation tasks. Biased estimates of the grasp pose cause undesired interaction wrenches on the human partner, which disturbs the interaction and the recognition of motion intention. A use of wearable motion sensors for tracking the human motion facilitates the grasp pose estimation without a global sensing system. This paper presents an approach for estimating an unknown grasp pose of the human using wearable motion sensors while minimizing undesired interaction wrenches applied to the human. A condition necessary for convergence of the estimator together with appropriate robot motion strategies are provided. Estimation of relative orientation and displacement is performed online and based on minimizing the error in the least-square sense. The estimation process does not rely on a global sensing system and it considers only the measurements of the velocity and acceleration of the cooperating partners in their respective local frames. The approach is experimentally evaluated in a physical human-robot interaction scenario.

In this paper, we present a method to plan grasps for soft hands. Considering that soft hands can easily conform to the shape an the object, with preference to certain types of basic geometries and dimensions, we decompose the object into one type of these geometries, particularly into Minimal Volume Bounding Boxes (MVBBs), which are proved to be efficiently graspable by the hand we use. A set of hand poses are then generated using geometric information extracted from such MVBBs. All hand postures are used in a dynamic simulator of the PISA/IIT Soft Hand and put on a test to evaluate if a proposed hand posture leads to a successful grasp. We show, through a set of numerical simulations, that the probability of success of the hand poses generated with the proposed algorithm is very good and represents an evident improvement with respect to our previous results published in [1].

We present an effective method of identifying the graspable components of a variety of complex objects for grasp planning based on human experience. Instead of focusing on individual objects, our method identifies graspable components on the category level under the assumption that geometrically alike objects share similar graspable components. Employing a modified SHOT descriptor, we propose a fast KNN-based method for object categorization. Then the graspable components are identified by adopting a learning framework based on human experience. Finally, a fast grasp planning method comprised of contact points exaction and hand kinematics calculation accomplishes the grasp on the identified graspable component. Our experiments demonstrate the effectiveness of this method by realizing grasps on the graspable components of human choice for a variety of unseen objects.

Bipedal walking is a complex and dynamic whole-body motion with balance constraints. Due to the inherently unstable inverted pendulum-like dynamics of walking, the design of robust walking controllers proved to be particularly challenging. While a controller could potentially be learned with a robot in the loop, the destructive nature of losing balance and the impracticality of a high number of repetitions render most existing learning methods unsuitable for an online learning setting with real hardware. We propose a model-driven learning method that enables a humanoid robot to quickly learn how to maintain its balance. We bootstrap the learning process with a central pattern generator for stepping motions that abstracts from the complexity of the walking motion and simplifies the problem setting to the learning of a small number of leg swing amplitude parameters. A simple physical model that represents the dominant dynamics of bipedal walking estimates an approximate gradient and suggests how to modify the swing amplitude to restore balance. In experiments with a real robot, we show that only a few failed steps are sufficient for our biped to learn strong push recovery skills in the sagittal direction.

In this paper, we present a new automatic camera control to achieve a relevant information framing. This camera control will be performed using a visual servoing framework in order to reach a salient area in a plane and in space. The relevant framing will be modelled by maximizing the saliency-based Gaussian Mixture Model (GMM) feature in the image. Furthermore, in order to achieve a realistic automatic camera control, we add an obstacles avoidance constraint and we ensure a relevant orientation during the motion. We validate our contribution in different synthetic 2D and 3D environments. Finally, we test our approach on a dense 3D points cloud model and in a real environment with a robot.

Global localization is a widely studied problem, and in essence corresponds to the online robot pose estimation based on a given map with landmarks, an odometry model, and real robot sensory observations and motion. In most approaches, the map provides the position of visible objects, which are then recognized to provide the robot pose estimation. Such object recognition with noisy sensory data is challenging. In this paper, we present an effective global localization technique using soft 3D object recognition to estimate the pose with respect to the landmarks in the given map. A depth sensor acquires a partial view for each observed object, from which our algorithm extracts the robot pose relative to the objects, based on a library of 3D Partial View Heat Kernel descriptors. Our approach departs from methods that require classification and registration against complete 3D models, which are prone to errors due to noisy sensory data and object misclassifications in the recognition stage. We experimentally validate our method in different robot paths with different common 3D environment objects. We also show the improvement of our method compared to when the partial view information is not used.

Sensor fusion combines multiple sensor measurements to improve a controller's knowledge about the internal state of an observed physical environment. Many such sensor fusion techniques exist and have been implemented for the Robot Operating System (ROS). However, they often have been developed for specific applications and cannot be easily reused for other applications. Reasons are the use of application-specific, partly undocumented interfaces, and the often limited reconfigurability caused by a tight coupling of the implementation to an application-specific purpose. Our approach is based on the concept of a fusion node which provides a configurable sensor fusion service with a generic interface. Fusion nodes can be interconnected to combine several sensor fusion techniques, can be attached to any single-dimension value sensor, can handle asynchronous multi-rate measurements and are robust regarding indeterministic, best-effort communication. This paper presents, to the best of our knowledge, the first generic sensor fusion package (GSFP) for ROS which collects various exemplary sensor fusion methods implemented as fusion nodes. We demonstrate the feasibility of our package in a small test application. Main benefits of our contribution are the developed ROS package's independence regarding specific sensors or applications, the easy integration of configurable fusion nodes in existing applications, and the composition of fusion nodes to realize complex sensor fusion scenarios.

This paper addresses the challenging issue of real-time camera localization relative to any object that have texture or not, sharp edges or occluding contours. 3D contour points, dynamically extracted from a CAD model by Analysis-by-Synthesis on the graphics hardware, are combined with a keyframe-based SLAM algorithm to estimate camera poses. Our tracking solution is accurate, robust to sudden motions and to occlusions, as demonstrated on synthetic and real data. This solution is also easy to deploy since it only uses an RGB camera and a CAD model of the object of interest, requires no manual intervention on this model and runs on a consumer tablet at a frequency of 40Hz on a HD video-stream. Videos are available as supplemental material.

Recently micro manipulation techniques for cell or microorganism has attracted attention in life sciences. In this paper, non-contact manipulation method using water stream is proposed. Water stream is generated by a glass needle vibration that is actuated by a piezoelectric actuator. The stream is controlled by changing input voltage or frequency to the actuator. An accuracy of the motion of a piezoelectric actuator is in nano-meter order, so the stream is considered to be controlled precisely. The manipulator works in 3 dimensional space, that means water stream is generated in 3d space. We focused on the rotational stream generated by the glass needle 2-dimensional circular vibration. By using the resonance in a structure of the manipulator and the actuator's vibration, circular vibration can be generated by only one actuator. There is no need to use two actuators for generating circular vibration, so the structure of the manipulator is quite simple. In this paper rotational stream around the glass needle is also analyzed in 3d space using micro beads. This manipulator has the similar characteristics to the existing research of contact manipulation, so we think our method also have the capability to conduct contact manipulation.

This paper considers the problem of generating dynamically feasible and collision free trajectory for unmanned aerial vehicles(UAVs) in cluttered environments. General random-based searching algorithms output piecewise linear paths, which cause big discrepancy when used as navigation reference for UAVs with high speed. Meanwhile, the disturbance may also occur to lead the UAVs into danger. In order to obtain agile autonomy without potential dangers, this paper introduces a three-step method to generate feasible reference. In the first step, a six-order Bezier curve, which uses Tuning Rotation to decrease the curvature, is introduced to smooth the output of the path planner. Then a forward simulation is implemented to find the potential dangerous regions. Finally, the path is reshaped by local optimal reshaping planner to eliminate residual dangers. The three steps form a circulation, the reshaped path sent to the first step again to check dynamic feasibility and safety. The method combining Six-order Bezier curve, Tuning Rotation, and local optimal reshaping is proposed by us for the first time, where the Tuning Rotation is able to meet various curvature requirements without violating the previous path, local optimal reshaping obtains both temporal and spatial reshaping with high time efficiency. The method addresses the system dynamics to achieve agile autonomy, which provides the geometry reference as well as the low level control. The effectiveness of the proposed method is demonstrated by the simulations.