IROS 2016 Papers

Series Elastic Actuators (SEAs) have several superiorities over conventional stiff and non-back-drivable actuators in force control, e.g., lower reflected inertia, low cost force measurement, high force fidelity, safety, and so on. However, their position control applications significantly suffer from low performance and disturbances due to insufficient controller designs. In this paper, a new Active Disturbance Rejection (ADR) controller is proposed for the robust position control problem of SEAs by combining Differential Flatness (DF) and Disturbance Observer (DOb) in state space. The trajectory of the actuator is generated by using DF and is tracked by using a conventional state feed-back controller. The state and control input references of a DF-based trajectory tracking controller are modified by using estimated disturbances so that the robustness is achieved. The proposed controller provides high performance position control for SEAs when they suffer from plant uncertainties and external disturbances such as inertia variation, backlash, friction and external load. Experimental results are given to validate the proposal.

This paper presents an active, model-based recognition system. It applies information theoretic measures in a belief-driven planning framework to recognize objects using the history of visual and manual interactions and to select the most informative actions. A generalization of the aspect graph is used to construct forward models of objects that account for visual transitions. We use populations of these models to define the belief state of the recognition problem. This paper focuses on the impact of the belief-space and object model representations on recognition efficiency and performance. A benchmarking system is introduced to execute controlled experiments in a challenging mobile manipulation domain. It offers a large population of objects that remain ambiguous from single sensor geometry or from visual or manual actions alone. Results are presented for recognition performance on this dataset using locomotive, pushing, and lifting controllers as the basis for active information gathering on single objects. An information theoretic approach that is greedy over the expected information gain is used to select informative actions, and its performance is compared to a sequence of random actions.

We describe an implementation of torsional work at high altitude employing an aerial robot equipped with an upward directed hand on top of a hexarotor airframe, considering its workspace above the airframe. The aerial robot in this study consists of a small hexarotor platform and a robotic hand module including a gripper. After grasping the object, the torsional work is executed by using yaw rotation of the body of the robot. For increasing a torque for the torsional task, an impact device is embedded at the base of the hand module. A maximum instant torque generated with this impact device was more than 4 N·m. In addition, the number of rotations of the body is measured through images captured by an onboard camera mounted beside the hand. Through experiments, we verified the feasibility of the present robot by successfully unscrewing a light bulb set to the socket above the robot and safely landing with the detached bulb.

We present the dynamic modeling, analysis, and control design of a Planar-Vertical Take-Off and Landing (PVTOL) underactuated aerial vehicle equipped either with a rigid- or an elastic-joint arm. We prove that in both cases the system is exactly linearizable with a dynamic feedback and differentially flat for the same set of outputs (but different controllers). We compare the two cases with extensive and realistic simulations, which show that the rigid-joint case outperforms the elastic-joint case for aerial grasping tasks while the converse holds for link-velocity amplification tasks. We present preliminary experimental results using a actuated joint with variable stiffness (VSA) on a quadrotor platform.

In this paper we consider a problem of delayed bilateral teleoperation control. We exploit a recent parameterization of all feasible teleoperator admittance matrices to propose a control synthesis procedure based on admittance shaping. This leads to an intuitive performance-oriented controller design that reveals important trade-offs inherent to bilateral teleoperaton. In particular, this allows a trade-off between performance and passivity, that is not accessible while using classical passivity-based methods. The proposed controller design is illustrated with simulations and a simple 1 DOF experimental case study.

This paper proposes a new adaptive repetitive visual-servo control system for a moving high-flying vehicle (HFV) with an uncalibrated camera to monitor, track, and precisely control the movements of a low-flying vehicle (LFV) or mobile ground robot. When deployed, a remote operator of the HFV defines the desired trajectory for the LFV in the HFV's camera frame (image frame). Due to the circulatory motion of the HFV, the resulting motion trajectory of the LFV in the image frame is periodic in time, thus an adaptive repetitive control system is exploited to improve the tracking precision from one operating period to the next. Not only is the adaptive control law able to deal with uncertainties in the camera's intrinsic and extrinsic parameters, but it can also tolerate uncertainties in the localization of the LFV. The design and stability analysis of the closed-loop control system is presented, where the Lyapunov approach is used to show stability. Simulations and experimental results are presented to demonstrate the effectiveness of the method for controlling the movement of a low-flying quadcopter vehicle. Results show good tracking performance for three simulated test cases, where the average maximum tracking error is reduced by approximately 75% compared to the performance of a standard adaptive visual-servo controller.

In this paper, a single mobile robot is used to patrol an indoor environment composed of rooms. The robot is required to monitor the rooms and detect visitors as many as possible. A key idea in the patrolling mission is to allow the robot to utilize information about visitor trends. Thus far, we have focused on a visiting probability as the trend. However, the visitor trends usually change over time. Therefore, the robot is further required to detect the changes and identify the correct visitor trends in each room. For this challenge, we present a novel change-point detection method. In this method, the Jensen-Shannon divergence is used as a dissimilarity measure between two distributions of the visiting probability. The change-point detection method is activated in the Bayesian learning for estimating the changed visiting probability. Consequently, the robot is allowed to adaptively patrol the changing environment. Through the simulation experiments, we discuss the effectiveness of the adaptive patrolling for the changes in the visitor trends.

Concentric tube robots comprise telescopic precurved elastic tubes. The robot's tip and shape are controlled via relative tube motions, i.e. tube rotations and translations. Non-linear interactions between the tubes, e.g. friction and torsion, as well as uncertainty in the physical properties of the tubes themselves, e.g. the Young's modulus, curvature, or stiffness, hinder accurate kinematic modelling. In this paper, we present a machine-learning-based methodology for kinematic modelling of concentric tube robots and in situ model adaptation. Our approach is based on Locally Weighted Projection Regression (LWPR). The model comprises an ensemble of linear models, each of which locally approximates the original complex kinematic relation. LWPR can accommodate for model deviations by adjusting the respective local models at run-time, resulting in an adaptive kinematics framework. We evaluated our approach on data gathered from a three-tube robot, and report high accuracy across the robot's configuration space.

Locomotion on the rough and variable ground surfaces is crucial for robots to complete various tasks. Recent advancement in numerical computation allow such locomotive robots to manipulate in real environments using a model-based control framework. This approach is successful if the precise model it is obtained. However, it is not always feasible to use the precise model because there are various factors related to both the robot and its surroundings. In this paper, we report the locomotion experiments with 2 types of legged robot that can change its behavior by switching connection patterns among hydraulic cylinders mounted on the robot's legs. An actuator network system (ANS) is used to switch mutual interconnection of cylinders. The switching connection allows the robots to realize not only adaptive locomotion to the environment but also operation of the traveling direction.

In this paper, we address the problem of the image-based 3D pose computation of a semi-circle suturing needle using monocular image feedback for laparoscopy. We propose a constrained two-degree-of-freedom (2-DOF) geometry-based modelling method to parametrise the needle's 6-DOF pose, including depth information. The modelling solely relies on the simultaneous observation of the needle's apparent tip and junction. No external markers are needed for extra constraints. An adaptive controller combining gradient descent and vector-flow method is introduced to iteratively guide the needle's initial guessing pose to its real pose by minimizing image-based position errors. Experiments have been conducted using both numerical simulations and simulated laparoscopic scenarios to evaluate the performance of the algorithm.

How should a robot direct active vision so as to ensure reliable grasping? We answer this question for the case of dexterous grasping of unfamiliar objects. By dexterous grasping we simply mean grasping by any hand with more than two fingers, such that the robot has some choice about where to place each finger. Such grasps typically fail in one of two ways, either unmodeled objects in the scene cause collisions or object reconstruction is insufficient to ensure that the grasp points provide a stable force closure. These problems can be solved more easily if active sensing is guided by the anticipated actions. Our approach has three stages. First, we take a single view and generate candidate grasps from the resulting partial object reconstruction. Second, we drive the active vision approach to maximise surface reconstruction quality around the planned contact points. During this phase, the anticipated grasp is continually refined. Third, we direct gaze to improve the safety of the planned reach to grasp trajectory. We show, on a dexterous manipulator with a camera on the wrist, that our approach (80.4% success rate) outperforms a randomised algorithm (64.3% success rate).

Accurate object shape knowledge provides important information for performing stable grasping and dexterous manipulation. When modeling an object using tactile sensors, touching the object surface at a fixed grid of points can be sample inefficient. In this paper, we present an active touch strategy to efficiently reduce the surface geometry uncertainty by leveraging a probabilistic representation of object surface. In particular, we model the object surface using a Gaussian process and use the associated uncertainty information to efficiently determine the next point to explore. We validate the resulting method for tactile object surface modeling using a real robot to reconstruct multiple, complex object surfaces.

In this paper, a new task-oriented active-sensing method is presented. Most active sensing methods choose sensing actions that minimize the uncertainty of the state according to some information-theoretic measure. While this is reasonable for most applications, minimizing state uncertainty may not be most relevant when the state information is used to perform a task. This is because the uncertainty in some subspace of the state space could have more impact on the performance of the task than the others at a given time. The active-sensing method presented in this paper takes the task into account when selecting sensing actions by minimizing the uncertainty in future task action.

Existing Simultaneous Localization and Mapping systems require an extensive manual pre-calibration process. Non-manual calibration procedures use manipulators to create known patterns in order to estimate the unknown calibration. Calibration is often time-consuming and involves humans performing repetitive tasks such as aligning a known calibration target at different poses with respect to the sensor. We propose an algorithm that plans a trajectory which actively reduces the uncertainty of the robot's calibration given a rough initial calibration estimate. Calibration is performed autonomously in a previously unknown environment by maintaining the belief over landmarks, poses, and the calibration parameters. We present experimental results to demonstrate the approach's ability to autonomously calibrate the exteroceptive sensor in simulated and real environments. We show that even a greedy approach can reduce the effort needed to perform calibration every time the robot is reconfigured for autonomous tasks and mitigates the possibility of human error added into the calibration.

This paper addresses the problem of robot global localization in a known environment, in the presence of many dynamic obstacles. Deploying a robot in crowded spaces such as museums, shopping malls, department stores, or university campuses is especially challenging because the moving people occlude the static parts of the environment, such as walls and doorways, making the robot essentially blind. A new weighting function is proposed for a particle filter state estimation algorithm that accounts for the presence of dynamic obstacles and avoids population depletion. An active localization strategy is employed which guides the robot to locations that resolve ambiguities and eliminate hypotheses in a systematic manner. Experimental results from multiple simulations and from real robot deployments validate the localization improvements achieved by the proposed method.

In this work we study the problem of exploring surfaces and building compact 3D representations of the environment surrounding a robot through active perception. We propose an online probabilistic framework that merges visual and tactile measurements using Gaussian Random Field and Gaussian Process Implicit Surfaces. The system investigates incomplete point clouds in order to find a small set of regions of interest which are then physically explored with a robotic arm equipped with tactile sensors. We show experimental results obtained using a PrimeSense camera, a Kinova Jaco2 robotic arm and Optoforce sensors on different scenarios. We then demostrate how to use the online framework for object detection and terrain classification.

In this paper, we propose a novel decentralized active perception strategy that maximizes the convergence rate in estimating the (unmeasurable) formation scale in the context of bearing-based formation localization for robots evolving in ℝ3 × S1. The proposed algorithm does not assume presence of a global reference frame and only requires bearing-rigidity of the formation (for the localization problem to admit a unique solution), and presence of (at least) one pair of robots in mutual visibility. Two different scenarios are considered in which the active scale estimation problem is treated either as a primary task or as a secondary objective with respect to the constraint of attaining a desired bearing formation. The theoretical results are validated by realistic simulations.

Many neurological or orthopedic disorders may cause motor impairments of shoulder. Patients require intensive training in order to recover from those impairments. However, the intensive training will lead to growing demand for strenuous work. Robot-aided therapy is able to alleviate the therapist's laborious burden and to provide more information about patient's condition during the therapy. Generally, the requirement of active therapy arises during the later stage of recovery. To realize the active therapy, a useful active control of the robot and effective recovery assessment for the active robot-aided therapy are essential. In this paper, a new active control method and an active therapy protocol with assessment indexes are designed. In this pilot research, several experiments on healthy subjects are designed to verify the proposed method with its active control performance and the effectiveness of the assessment indexes.

Active Constraint Learning (ACL) is continuously gaining popularity in the area of constrained clustering due to its ability to achieve performance gains via incorporating minimal feedback from a human annotator for selected instances. For constrained clustering algorithms, such instances are integrated in the form of Must-Link (ML) and Cannot-Link (CL) constraints. Existing iterative uncertainty reduction schemes, introduce high computational burden particularly when they process larger datasets that are usually present in computer vision and visual learning applications. For scenarios that multiple agents (i.e., robots) require user feedback for performing recognition tasks, minimizing the interaction between the user and the agents, without compromising performance, is an essential task. In this study, a non-iterative ACL scheme with proven performance benefits is presented. We select to demonstrate the effectiveness of our methodology by building on the well known K-Means algorithm for clustering; one can easily extend it to alternative clustering schemes. The proposed methodology introduces the use of the Silhouette values, conventionally used for measuring clustering performance, in order to rank the degree of information content of the various samples. In addition, an efficient greedy selection scheme was devised for selecting the most informative samples for human annotation. To the best of our knowledge, this is the first active constrained clustering methodology with the ability to process computer vision datasets that this study targets. Performance results are shown on various computer vision benchmarks and support the merits of adopting the proposed scheme.