ICRA 2017 Papers

In-hand manipulation is certainly one of the most challenging problems in robotic manipulation. Solutions to this problem depend on the specific device used to grab the object, but nowadays, the trend is to exploit not only the gripper but also external constraints, such as other objects in the environment or external forces, like gravity. This allows a robot to manipulate an object even with very simple grippers, like a parallel gripper. Nevertheless, even for a simple grasping task, which aims at grabbing the object with a given fixed orientation or for executing a controlled slip, information on the contact between the fingers of the gripper and the object is relevant. In these cases, both linear and rotational slipping should be controlled during the grasping phase and during the motion phase. The present paper proposes a control strategy for the first objective, namely slipping avoidance. The strategy is based on contact information provided by a six-axis force/tactile sensor, able to measure contact force and torque as well as able to provide information on the contact geometry, that means orientation of the object with respect to the gripper. Experiments on a parallel gripper sensorized with a new force/tactile sensor and mounted on a Kuka iiwa show how the strategy successfully allows the robot to safely manipulate a rigid object in various friction conditions of its surface.

This work explores the stabilization of desired dynamic motion tasks involving hard impacts at non-negligible speed for humanoid robots. To this end, a so-called reference spreading hybrid control law is designed showing promising results in simulation. The simulations are performed employing a dynamical model of an existing humanoid robot and impacts are assumed to be inelastic. The desired motion task consists of having the robot balancing on one foot while repeatedly making and breaking contact with a wall by means of one hand. The simulation results illustrate that the considered controller is suited to control humanoid robot motions with impacts.

Bio-syncretic robots, consisting of living biological materials and traditional electromechanical systems, have attracted lots of attention due to the potentialities of self-sensing, self-actuation and self-repairing with intrinsic safety and high energy conversion efficiency. However, most of current researches focus on the movement of the devices, and have ignored the study on the control of actuation unit “cells”, which is as important as motors for traditional electromechanical robots. In this work, the effects of cell culturing time, seeding concentration and functional drugs (cytochalasin and adrenalin) on contractile frequency and force strength of cardiomyocytes have been studied using scanning ion conductance microscope (SICM) and arrays of micro-pillars made of PDMS. This work will lay the foundation for the further study of quantitatively control of bio-syncretic robots actuated by cardiomyocytes and is also meaningful for the development of cytology, medicine, and clinical science.

Humanlike robot skills, e.g., cleaning a table or handing over a plate, can often be generalized to different task variations. Usually, these are start-/goal position, and trained environment changes. We investigate how to modify motion primitives to context changes, which are not included in the training data. Specifically, we focus on maintaining humanlike motion characteristics and generalizability, while adapting to unseen context. Therefore, we present an optimization technique, which maximizes the expected return and minimizes the Kullback-Leibler Divergence to the demonstrations at the same time. Simultaneously, our algorithm learns how to linearly combine the adapted primitive with the demonstrations, such that only relevant parts of the primitive are adapted. We evaluate our approach in obstacle avoidance and broken joint scenarios in simulation, as well as on a real robot.

In fast changing assembly scenarios, it is required to adapt the task execution to the current state of the setup without extensive calibration routines. Therefore, it is important to estimate the geometric uncertainties and contact states during the assembly execution. We use a sequential Monte Carlo (SMC) method to track the relative poses between workpieces during a robotic assembly based on joint torque and position measurements only. In contrast to existing approaches, we focus on assembly tasks where the workpiece is not fixed in the workcell, but can, for example, slide on a table surface. We propose a new constraint-based propagation model for the SMC approach: a compensation motion for the samples dependent on the violation of contact constraints is derived. This allows us to track the motion of the workpieces in cases where a common random diffusion model fails. The method is evaluated with experiments using an assembly scenario with two KUKA LBR iiwa robot arms and shows accurate tracking performance.

In this paper, we address the problem of how a robot can optimize parameters of combined interaction force/task space controllers under a success constraint in an active way. To enable the robot to explore its environment robustly, safely and without the risk of damaging anything, suitable control concepts have to be developed that enable compliant and force control in situations that are afflicted with high uncertainties. Instances of such concepts are impedance, operational space or hybrid control. However, the parameters of these controllers have to be tuned precisely in order to achieve reasonable performance, which is inherently challenging, as often no sufficient model of the environment is available. To overcome this, we propose to use constrained Bayesian optimization to enable the robot to tune its controller parameters autonomously. Unlike other controller tuning methods, this method allows us to include a success constraint into the optimization. Further, we introduce novel performance measures for compliant, force controlled robots. In real world experiments we show that our approach is able to optimize the parameters for a task that consists of establishing and maintaining contact between the robot and the environment efficiently and successfully.

In this paper we introduce a novel approach for efficient decision making under uncertainty and belief space planning, in high dimensional state spaces. While recently developed methods focus on sparsifying the inference process, the sparsification here is done in the context of efficient decision making, with no impact on the state inference. By identifying state variables which are uninvolved in the decision, we generate a sparse version of the state's information matrix, to be used in the examination of candidate actions. This sparse approximation is action-consistent, i.e. has no influence on the action selection. Overall we manage to maintain the same quality of solution, while reducing the computational complexity of the problem. The approach is put to the test in a SLAM simulation, where a significant improvement in runtime is achieved. Nevertheless, the method is generic, and not tied to a specific type of problem.

In this paper, we seek to provide consistent, real-time 3D localization capabilities to mobile devices navigating within previously mapped areas. To this end, we introduce the Cholesky-Schmidt-Kalman filter (C-SKF), which explicitly considers the uncertainty of the prior map, by employing the sparse Cholesky factor of the map's Hessian, instead of its dense covariance-as is the case for the Schmidt-Kalman filter. By doing so, the C-SKF has memory requirements typically linear in the size of the map, as opposed to quadratic for storing the map's covariance. Moreover, and in order to bound the processing needs of the C-SKF (between linear and quadratic in the size of the map), we introduce two relaxations of the C-SKF algorithm: (i) The sC-SKF, which operates on the Cholesky factors of independent sub-maps resulting from dividing the map into overlapping segments. (ii) We formulate an efficient method for sparsifying the Cholesky factor by selecting and processing a subset of loop-closure measurements based on their temporal distribution. Lastly, we assess the processing and memory requirements of the proposed algorithms, and compare their positioning accuracy against other inconsistent map-based localization approaches that employ measurement-noise-covariance inflation to compensate for the map's uncertainty.

This paper solves the problem of computing the best grasp in a discrete point set based on a popular grasp quality measure, namely the largest origin-centered ball contained in the grasp wrench set. So far, the solution to this problem is very limited. Noticing that the quality measure for a grasp is equal to the minimum value of the support function of its grasp wrench set over all directions and its computation together with force closure test can be fulfilled by evaluating the support function in a sequence of directions, we can quickly determine that a new grasp is worse whenever its support function in the sequence of directions or any other specific direction is less than the quality value of the current best grasp and avoid further computation. Furthermore, we enumerate candidate grasps in the point set in an adaptive way such that grasps that are more likely to outperform the current best grasp will be checked first, which helps find the best grasp earlier and significantly reduce the number of candidate grasps to be fully examined. With the aid of the adaptive enumeration and the quick comparison of grasps, the proposed algorithm takes tens of seconds to several hours on a normal PC to compute the best grasp in tens to hundreds of points on 3-D objects and it is two to several orders of magnitude faster than the brute-force search.

We present a computational design system that allows novices and experts alike to easily create custom robotic devices using modular electromechanical components. The core of our work consists of a design abstraction that models the way in which these components can be combined to form complex robotic systems. We use this abstraction to develop a visual design environment that enables an intuitive exploration of the space of robots that can be created using a given set of actuators, mounting brackets and 3d-printable components. Our computational system also provides support for design auto-completion operations, which further simplifies the task of creating robotic devices. Once robot designs are finished, they can be tested in physical simulations and iteratively improved until they meet the individual needs of their users. We demonstrate the versatility of our computational design system by creating an assortment of legged and wheeled robotic devices. To test the physical feasibility of our designs, we fabricate a wheeled device equipped with a 5-DOF arm and a quadrupedal robot.

Currently, Compressive Tracking (CT) method has drawn great attention because of its high efficiency. However, it cannot well deal with some appearance variations due to its limitations of feature expression and it only uses a fixed parameter to update the appearance model. In order to handle such matters, we propose an adaptive CT method that combines the predicted target position with CT based on Locality Sensitive Histograms (LSH) features. Our method significantly improves CT in four aspects. First, the efficient illumination invariant features extracted based on LSH are used to represent an effective appearance model that is robust to illumination changes. Second, the color attributes tracker is adopted to predict the target position for re-building the new weighted discriminant function which brings in the color information to make up for the inadequacy of Haar-like characteristics. Third, a new model update mechanism is proposed to preserve the stable features while avoid the noisy appearance variations during tracking. Fourth, a trajectory rectification method is employed to refine the tracking location when possible inaccurate tracking occurs. Finally, we show that our tracker achieves state-of-the-art performance in a comprehensive evaluation over 47 challenging color sequences.

In this work, a clustering approach to obtain compact topological models of an environment is developed and evaluated. The usefulness of these models is tested by studying their utility to solve the robot localization problem subsequently. Omnidirectional visual information and global appearance descriptors are used both to create and compress the models and to estimate the position of the robot. Comparing to the methods based on the extraction and description of landmarks, global appearance approaches permit building models that can be handled and interpreted more intuitively and using relatively straightforward algorithms to estimate the position of the robot. The proposed algorithms are tested with a set of panoramic images captured with a catadioptric vision sensor in a large environment under real working conditions. The results show that it is possible to compress substantially the visual information contained in topological models to arrive to a balance between the computational cost and the accuracy of the localization process.

The ability to change stiffness is a capability exhibited through the animal kingdom, with many recent advances in tunable stiffness in the area of robotics. In this paper, we propose a mechanism design that provides the ability to make modular subcomponents with tunable stiffness by creating a bi-stable mechanism that exhibits different stiffnesses in each of the stable configurations. The design is based on controlled buckling of two linear springs in series and allows design-time control over the stiffnesses, equilibrium points, and energy required to transition between the stable configurations.

Assist-as-needed control aims at maximizing stroke survivors involvement during robotic-led therapies of neurorehabilitation. Besides the specific characteristics of the designed adaptive control strategy, a fundamental property of this control architecture is the choice of the error signal which will drive the adaptation process. This driving source is a necessary control parameter to be chosen, although often sidelined in the control design, and several solutions already exist in the state-of-the-art. For this reason, we wanted to compare three different strategies to guide the adaptation, respectively based on the local joint performances, on the end-effector only behaviour, or on the performance of one specific joint in the kinematic chain of the robot. The resulting analysis evaluates the possibilities offered by simply changing from one source to another with respect to the specific stage of the motor recovery of the patients, potentially extending the capabilities of current exoskeleton controllers for neurorehabilitation.

Motivated by recent advances in Deep Learning for robot control, this paper considers two learning algorithms in terms of how they acquire demonstrations from fallible human supervisors. Human-Centric (HC) sampling is a standard supervised learning algorithm, where a human supervisor demonstrates the task by teleoperating the robot to provide trajectories consisting of state-control pairs. Robot-Centric (RC) sampling is an increasingly popular alternative used in algorithms such as DAgger, where a human supervisor observes the robot execute a learned policy and provides corrective control labels for each state visited. We suggest RC sampling can be challenging for human supervisors and prone to mislabeling. RC sampling can also induce error in policy performance because it repeatedly visits areas of the state space that are harder to learn. Although policies learned with RC sampling can be superior to HC sampling for standard learning models such as linear SVMs, policies learned with HC sampling may be comparable to RC when applied to expressive learning models such as deep learning and hyper-parametric decision trees, which can achieve very low training error provided there is enough data. We compare HC and RC using a grid world environment and a physical robot singulation task. In the latter the input is a binary image of objects on a planar worksurface and the policy generates a motion in the gripper to separate one object from the rest. We observe in simulation that for linear SVMs, policies learned with RC outperformed those learned with HC but that using deep models this advantage disappears. We also find that with RC, the corrective control labels provided by humans can be highly inconsistent. We prove there exists a class of examples in which at the limit, HC is guaranteed to converge to an optimal policy while RC may fail to converge. These results suggest a form of HC sampling may be preferable for highly-expressive learning models and human supervisors. Show More

Tracking human gait accurately and robustly constitutes a key factor for a smart robotic walker, aiming to provide assistance to patients with different mobility impairment. A context-aware assistive robot needs constant knowledge of the user's kinematic state to assess the gait status and adjust its movement properly to provide optimal assistance. In this work, we experimentally validate the performance of two gait tracking algorithms using data from elderly patients; the first algorithm employs a Kalman Filter (KF), while the second one tracks the user legs separately using two probabilistically associated Particle Filters (PFs). The algorithms are compared according to their accuracy and robustness, using data captured from real experiments, where elderly subjects performed specific walking scenarios with physical assistance from a prototype Robotic Rollator. Sensorial data were provided by a laser rangefinder mounted on the robotic platform recording the movement of the user's legs. The accuracy of the proposed algorithms is analysed and validated with respect to ground truth data provided by a Motion Capture system tracking a set of visual markers worn by the patients. The robustness of the two tracking algorithms is also analysed comparatively in a complex maneuvering scenario. Current experimental findings demonstrate the superior performance of the PFs in difficult cases of occlusions and clutter, where KF tracking often fails.

Manipulation of deformable objects, such as ropes and cloth, is an important but challenging problem in robotics. We present a learning-based system where a robot takes as input a sequence of images of a human manipulating a rope from an initial to goal configuration, and outputs a sequence of actions that can reproduce the human demonstration, using only monocular images as input. To perform this task, the robot learns a pixel-level inverse dynamics model of rope manipulation directly from images in a self-supervised manner, using about 60K interactions with the rope collected autonomously by the robot. The human demonstration provides a high-level plan of what to do and the low-level inverse model is used to execute the plan. We show that by combining the high and low-level plans, the robot can successfully manipulate a rope into a variety of target shapes using only a sequence of human-provided images for direction.

We consider the global tracking control problem of robots with elastic joints. Even if joint elasticity introduces beneficial features for modern applications which require physically resilient and safer robots that can interact with the environment or humans, it challenges the achievable control performance. We propose a novel controller which combines the benefits of two approaches: the intrinsic robustness to model uncertainty from passivity-based control and the implementation efficiency of inverse-dynamics control schemes using a modern recursive algorithm. The novel controller is applied to an elastic-joint reconfigurable robotic arm using a recently proposed framework for on-the-fly control design. Simulation and experimental results validate our proposed approach.

Tracking in urban street scenes plays a central role in autonomous systems such as self-driving cars. Most of the current vision-based tracking methods perform tracking in the image domain. Other approaches, e.g. based on LIDAR and radar, track purely in 3D. While some vision-based tracking methods invoke 3D information in parts of their pipeline, and some 3D-based methods utilize image-based information in components of their approach, we propose to use image- and world-space information jointly throughout our method. We present our tracking pipeline as a 3D extension of image-based tracking. From enhancing the detections with 3D measurements to the reported positions of every tracked object, we use world-space 3D information at every stage of processing. We accomplish this by our novel coupled 2D-3D Kalman filter, combined with a conceptually clean and extendable hypothesize-and-select framework. Our approach matches the current state-of-the-art on the official KITTI benchmark, which performs evaluation in the 2D image domain only. Further experiments show significant improvements in 3D localization precision by enabling our coupled 2D-3D tracking.