ICRA 2015 Papers

This is the first demonstration of a modular and reconfigurable magnetic-manipulation system with integral ferromagnetic material. This system-which includes multiple Omnimagnets, each comprising three orthogonal solenoids and a spherical ferromagnetic core-is capable of dexterous manipulation of a magnetic tool. The magnetization coupling of an arbitrary arrangement of spherical ferromagnetic cores is solved, enabling an analytical solution for the magnetic forces and torques. Thus, the system does not require extensive field maps or in situ field characterization. If the positions and orientations of the Omnimagnets are sensed, the Omnimagnets can be actively rearranged during manipulation while maintaining control of the tool. This could enable new capabilities in medical procedures because the manipulation system can be modified to accommodate the imaging systems and enable physician access without loss of control. The capabilities of this system are demonstrated through five-degree-of-freedom manipulation (position and heading) of a permanent-magnetic tool with several arbitrary Omnimagnet configurations.

This paper presents a bio-inspired robotic fish undulatory swimming behaviour modeling and control using modulated pattern generators (MPG). Carangiform fish locomotion pattern is closely mimicked using a LH body wave (with its parameters: Tail-beat frequency (TBF) and Caudal amplitude (CA)) modulated by rhythmic central pattern generator (CPG) signal. A Matsuoka based non-linear oscillator CPG structure is used to generate desired rhythmic pattern preserving control properties like system stability (limit cycle behaviour) and synchronization. A two level locomotion control architecture based on vertebrate fish biology is proposed. Higher level controller generates desired trajectory as encoded patterns MPGs. Synchronization of the desired trajectories at each joint with the robotic fish dynamics generates desired fishlike locomotion behavior. Lower level control scheme uses an inverse dynamics model based policy for tracking this locomotion pattern (joint positions). It is shown that MPGs can represent the shape (amplitude) and phase (frequency) patterns of high-dimensional periodic inputs using simple kinematic parameters. Proposed inverse dynamics model based tracking control strategy is found to perform in a stable operating margin..

Vibration-driven locomotion has been widely used for crawling robot studies. Such robots usually have a vibration motor as the actuator and a fibrillar structure for providing directional friction on the substrate. However, there has not been any studies about the effect of fiber structure on robot crawling performance. In this paper, we develop Fiberbot, a custom made mini vibration robot, for studying the effect of fiber angle on robot velocity, steering, and climbing performance. It is known that the friction force with and against fibers depends on the fiber angle. Thus, we first present a new fabrication method for making millimeter scale fibers at a wide range of angles. We then show that using 30° angle fibers that have the highest friction anisotropy (ratio of backward to forward friction force) among the other fibers we fabricated in this study, Fiberbot speed on glass increases to 13.8±0.4 cm/s (compared to ν = 0.6±0.1 cm/s using vertical fibers). We also demonstrate that the locomotion direction of Fiberbot depends on the tilting direction of fibers and we can steer the robot by rotating the fiber pad. Fiberbot could also climb on glass at inclinations of up to 10° when equipped with fibers of high friction anisotropy. We show that adding a rigid tail to the robot it can climb on glass at 25° inclines. Moreover, the robot is able to crawl on rough surfaces such as wood (ν = 10.0±0.2 cm/s using 30° fiber pad). Fiberbot, a low-cost vibration robot equipped with a custom-designed fiber pad with steering and climbing capabilities could be used for studies on collective behavior on a wide range of topographies as well as search and exploratory missions.

Soft pneumatic actuators provide many exciting properties, but controlling them without the use of bulky and expensive flow-control valves can be difficult and computationally expensive. We seek a solution to this problem by introducing an inexpensive and reliable muscle-like linear soft actuator used antagonistically to operate a rigid 1-DoF joint, resulting in a system that combines the advantages of rigid and soft robotics. Using this setup, we performed precise motion control using a sliding mode feedback controller as well as a sliding mode controller augmented by a feedforward term to modulate the state of solenoid valves that drive each actuator. We found that both controllers performed equivalently well in following a step function and in responding to a disturbance. The feedforward augmented controller performed significantly better when following dynamic trajectories over a range of frequencies and with the addition of an external force. The next step will be to modify our valve control scheme to allow for the determination of both the position and stiffness of the joint, better leveraging the advantages of soft pneumatic actuators.

In this paper, an ongoing work for the implementation of a variable stiffness joint actuated by a couple of twisted string actuators in antagonistic configuration is reported. The twisted string actuation system is particularly suitable for very compact and light-weight robotic devices, like artificial limbs, exoskeletons and robotic hands, since it renders a very low apparent inertia at the load side, allowing the implementation of powerful tendon-based driving systems, using as actuators small-size DC motors characterized by high speed, low torque and very limited inertia. The basic properties of the twisted string actuation system are firstly presented, and the way how they are exploited for the implementation of a variable stiffness joint is discussed. A simple control algorithm for controlling the joint stiffness and position simultaneously is discussed, and a the feedback linearization of the device is taken into account and validated in simulation.

In this paper we present a feed forward controller to regulate the depth of laser incisions in soft tissue. Such a controller is compatible with the requirements of laser microsurgery, where space constraints limit the use of sensing devices. The controller is based on an inverse model that maps the desired incision depth to the required laser exposure time. This model is extracted from experimental data through the use of statistical learning methods. To prove the concept, the controller is implemented in a robot-assisted laser microsurgery system that enables precision control of exposure time and laser motion. The validity and the accuracy of the controller is verified experimentally on ex-vivo muscle tissue (chicken breast), revealing an RMSE of 0.12 mm for incisions ranging up to 1 mm. In addition, we demonstrate how the model can be used to implement the automatic ablation of entire volumes of tissue, through the superposition of controlled laser incisions.

Locomotion stability of a humanoid robot is closely related to the capacity to regulate its Center of Mass (CoM) motion. In this paper, the Feasible Center of Mass Dynamic Manipulability (FCDM) is introduced and analyzed as a measure of this capacity. The effects of posture, joint velocities and gravity on the torque-bounded dynamic manipulability ellipsoid are first analyzed on an n-DOF planar humanoid robot with single-foot support. The ellipse orientation has a linear relationship with the ankle angle, and its shape is independent on the ankle angle. Furthermore, three common and important ground-contact constraints - the unilateral contact-force constraint, the friction constraint, and the Center of Pressure constraint - are incorporated in the derivation of FCDM. It shows geometrically how each of the three constraints shrinks the original torque-bounded manipulability polytope and affects the maximum achievable CoM acceleration in different directions. Finally, a push recovery task was simulated to show that a robot's posture affects the feasible range of the CoM acceleration in a specific direction.

A soft wearable robot with a tendon drive is a promising technology that enables a wearable robot to be compact and lightweight. A soft tendon routing system was previously proposed to apply a tendon drive to a soft wearable robot.

This paper presents a fault tolerant approach for a coaxial octorotor regarding rotor failures. A complete architecture including error detection, fault isolation and system recovery is presented. The diagnosis system is designed with a nonlinear observer to generate residuals and an inference model to evaluate them and isolate the faulty motor. Once the motor failure is diagnosed, a recovery algorithm is applied. It uses the built-in hardware redundancy of the octorotor and compensates the loss of the failing motor by controlling its dual to keep a stable flight that allows the multirotor to continue its mission. This architecture is validated on real flights.

Autonomous navigation in real-world industrial environments is a challenging task in many respects. One of the key open challenges is fast planning and execution of trajectories to reach arbitrary target positions and orientations with high accuracy and precision, while taking into account non-holonomic vehicle constraints. In recent years, lattice-based motion planners have been successfully used to generate kinematically and kinodynamically feasible motions for non-holonomic vehicles. However, the discretized nature of these algorithms induces discontinuities in both state and control space of the obtained trajectories, resulting in a mismatch between the achieved and the target end pose of the vehicle. As endpose accuracy is critical for the successful loading and unloading of cargo in typical industrial applications, automatically planned paths have not be widely adopted in commercial AGV systems. The main contribution of this paper addresses this shortcoming by introducing a path smoothing approach, which builds on the output of a lattice-based motion planner to generate smooth drivable trajectories for non-holonomic industrial vehicles. In real world tests presented in this paper we demonstrate that the proposed approach is fast enough for online use (it computes trajectories faster than they can be driven) and highly accurate. In 100 repetitions we achieve mean end-point pose errors below 0.01 meters in translation and 0.002 radians in orientation. Even the maximum errors are very small: only 0.02 meters in translation and 0.008 radians in orientation.

Interest in human-robot coexistence, in which humans and robots share a common work volume, is increasing in manufacturing environments. Efficient work coordination requires both awareness of the human pose and a plan of action for both human and robot agents in order to compute robot motion trajectories that synchronize naturally with human motion. In this paper, we present a data-driven approach that synthesizes anticipatory knowledge of both human motions and subsequent action steps in order to predict in real-time the intended target of a human performing a reaching motion. Motion-level anticipatory models are constructed using multiple demonstrations of human reaching motions. We produce a library of motions from human demonstrations, based on a statistical representation of the degrees of freedom of the human arm, using time series analysis, wherein each time step is encoded as a multivariate Gaussian distribution. We demonstrate the benefits of this approach through offline statistical analysis of human motion data. The results indicate a considerable improvement over prior techniques in early prediction, achieving 70% or higher correct classification on average for the first third of the trajectory (< 500msec). We also indicate proof-of-concept through the demonstration of a human-robot cooperative manipulation task performed with a PR2 robot. Finally, we analyze the quality of task-level anticipatory knowledge required to improve prediction performance early in the human motion trajectory.

In this paper, we present an efficient semantic segmentation framework for indoor scenes operating on 3D point clouds. We use the results of a Random Forest Classifier to initialize the unary potentials of a densely interconnected Conditional Random Field, for which we learn the parameters for the pairwise potentials from training data. These potentials capture and model common spatial relations between class labels, which can often be observed in indoor scenes. We evaluate our approach on the popular NYU Depth datasets, for which it achieves superior results compared to the current state of the art. Exploiting parallelization and applying an efficient CRF inference method based on mean field approximation, our framework is able to process full resolution Kinect point clouds in half a second on a regular laptop, more than twice as fast as comparable methods.

We present Partial Quick Hull (PQH), an algorithm to efficiently compute one of the most commonly used grasp quality metrics. The metric relies on the computation of the convex hull of a set of points in a six dimensional space. Built on top of widely used QuickHull algorithm, PQH exploits the relationship between the convex hull and the grasp quality metric to avoid computing the whole convex hull. PQH determines at run time when the computation can be ended because the grasp quality metric can be already determined from a partially computed convex hull - hence the name of the algorithm. This improvement greatly accelerates grasp quality evaluation for force closure grasps. When the grasp is not force closure, the partial computation does not apply and PQH then behaves exactly like QuickHull. A large set of experimental tests show how PQH largely outperforms QuickHull and better scales with the size of the input.

Assisted gait monitoring could benefit from the measurement of feet position and orientation. These measurements could also be used to implement intention-based control laws for smart walkers. Several works have been dedicated to the detection of lower limbs. The proposed methods are usually fast, but only detect the position of the legs. Others may even be more complete, but are not adapted to reactive applications. Also, they often use markers attached on the feet, which is unsuitable in daily routine. In this paper, a fast feet position and orientation detection algorithm is proposed. It is based on a camera depth sensor and does not require the use of any marker. The obtained results are compared with a ground truth provided by a motion traking system to experimentally assess the performances of the proposed algorithm.

The stereo correspondence problem is still a highly active topic of research with many applications in the robotic domain. Still many state of the art algorithms proposed to date are unable to reasonably handle high resolution images due to their run time complexities or memory requirements. In this work we propose a novel stereo correspondence estimation algorithm that employs binary locality sensitive hashing and is well suited to implementation on the GPU. Our proposed method is capable of processing very high-resolution stereo images at near real-time rates. An evaluation on the new Middlebury and Disney high-resolution stereo benchmarks demonstrates that our proposed method performs well compared to existing state of the art algorithms.

Many estimation problems in robotics rely on efficiently solving nonlinear least squares (NLS). For example, it is well known that the simultaneous localisation and mapping (SLAM) problem can be formulated as a maximum likelihood estimation (MLE) and solved using NLS, yielding a mean state vector. However, for many applications recovering only the mean vector is not enough. Data association, active decisions, next best view, are only few of the applications that require fast state covariance recovery. The problem is not simple since, in general, the covariance is obtained by inverting the system matrix and the result is dense. The main contribution of this paper is a novel algorithm for fast incremental covariance update, complemented by a highly efficient implementation of the covariance recovery. This combination yields to two orders of magnitude reduction in computation time, compared to the other state of the art solutions. The proposed algorithm is applicable to any NLS solver implementation, and does not depend on incremental strategies described in our previous papers, which are not a subject of this paper.

This paper presents a fast collision-detection method for sampling-based motion planners based on bounding volume hierarchies in workspace-time space. By introducing time as an additional dimension to the robot's workspace, the method is able to quickly evaluate time-indexed candidate trajectories for collision with the known future motions of other agents. The approach makes no assumptions on the shape of the objects and is able to handle arbitrary motions. We highlight implementation details regarding the application of the collision detection technique within an online planning framework for automated driving. Furthermore, we give detailed profiling information to show the capability for real-time operation.

In this paper, we propose a car positioning approach that does not rely on GPS. We propose to use car wheel speeds and road maps in order to achieve robust positioning of the vehicle. The vehicle positioning is achieved by applying particle filtering on a graph-based representation of a road map. We show that the vehicle positioning is feasible and robust with these two inputs at a really low computational cost. We achieve car positioning with an averaged 5 m accuracy within a 100 km drivable road map on a 12 km sequence.

Estimating the robot localization is a fundamental requirement for applications in robotics. For many years, Monte Carlo Localization (MCL) has been one of the most popular approaches to solve the global localization when using range finders, like sonars or lasers. It generally weights the estimates about the robot state by comparing raw sensor readings with simulated readings computed for each estimate. In this paper, we propose an observation model for localization that associates a kernel density estimate (KDE) to each point in the space. This single-valued density measure is independent of orientation, what allows an efficient pre-caching step, substantially boosting the computation time of the process. Using the gradient of the densities field, our strategy is able to estimate orientation information that helps to restrict the localization search space. Additionally, we can combine densities obtained by kernels of different sizes and profiles to improve the quality of the acquired information. We show through experiments in comparison with traditional approaches that our method is efficient, even working with large sets of particles, and effective.