IROS 2016 Papers

In this paper, a new type of an active-caster with a single ball transmission (ACROBAT-S) is proposed. The proposed ball transmission is a novel mechanism which realizes combining of two motor powers together to rotate the single ball in a 2Dway and simultaneously distributing the combined power to a wheel shaft and a steering shaft of an active-caster in an appropriate ratio. The transmission design enables to remove a sensor for detecting the wheel orientation for coordinated control. Also, it would be possible to build an omnidirectional robot with three active casters to be controlled its 3D motion by three motors with no redundancy. The new concept of ACROBAT-S gives many advantages compared with the original ACROBAT since the number of balls is reduced from two to one, therefore the number of friction drive between a ball and rollers or a ball to a ball is also reduced from five to three. These features contribute not only to simplify the mechanism design, but also to enhance its performance since slippages and energy losses would be reduced. To verify the design concept, we derive and analyze a kinematic model of a single ball transmission mechanism. Furthermore the motions of ACROBAT-S are verified by Solid Works 3D simulator. From the results, it is confirmed that the proposed single ball transmission is applicable for an omnidirectional wheel mechanism to realize omnidirectional motions.

This paper aims at providing a novel approach to modeling and controlling soft robots. Based on real-time Finite Element Method (FEM), we obtain a globally defined discrete-time kinematic model in the workspace of soft robots. From the kinematic equations, we deduce the soft-robot Jacobian matrix and discuss the conditions to avoid singular configurations. Then, we propose a novel observer based control methodology where the observer is built by Finite Element Model in this paper to deal with the control problem of soft robots. A closed-loop controller for position control of soft robot is designed based on the discrete-time model with feedback signal being extracted by means of visual servoing. Finally, experimental results on a parallel soft robot show the efficiency and performance of our proposed controller.

We introduce the Kilogrid, a modular and scalable virtualization environment aimed at swarm robotics research with the Kilobot robot. The main purpose of the Kilogrid is to complement the Kilobots by overcoming some of their limitations (i.e., limited sensors and actuators), making it easier to experiment and to collect data with large groups of robots. The Kilogrid allows researchers to study scenarios featuring a level of complexity that cannot be reached using the Kilobots alone. The Kilogrid is composed of several modules, where each module contains four cells of 50×50 mm2. The cells allow for bi-directional communication with the Kilobots. Our first version of a Kilogrid is composed of 64 cells and covers a total area of 400×400 mm2. We demonstrate the features of the Kilogrid with two case studies in which: (i) we extend the sensory system of the Kilobots, (ii) we allow the Kilobots to modify the environment, and (iii) we collect data (e.g., position, state) from the Kilobots while the experiment is running.

In this paper, we propose a keyframe-based online object learning and detection method. To manage appearance changes of target objects, the proposed method incrementally updates an object database using detection results. One of the major problems in updating the appearance model is that the object model can gradually be degraded by accumulated errors and biased to specific views. To solve this problem, our object model is updated according to the selected keyframes, which not only help memorize important views of target objects, but also prevent the holistic appearance model from overfitting. The database is represented as a graph of the registered images, and the importance of the database images is measured by analyzing the constructed graph. Then, the redundant or less important images are discarded from the database. As a result, the database is efficiently maintained while new views of the objects are gradually added. The experimental results demonstrate that the proposed algorithm efficiently maintains the object database and improves the detection performance compared to previous incremental object learning and detection algorithms.

This paper studies indoor localisation problem by using low-cost and pervasive sensors. Most of existing indoor localisation algorithms rely on camera, laser scanner, floor plan or other pre-installed infrastructure to achieve sub-meter or sub-centimetre localisation accuracy. However, in some circumstances these required devices or information may be unavailable or too expensive in terms of cost or deployment. This paper presents a novel keyframe based Pose Graph Simultaneous Localisation and Mapping (SLAM) method, which correlates ambient geomagnetic field with motion pattern and employs low-cost sensors commonly equipped in mobile devices, to provide positioning in both unknown and known environments. Extensive experiments are conducted in large-scale indoor environments to verify that the proposed method can achieve high localisation accuracy similar to state-of-the-arts, such as vision based Google Project Tango.

This paper introduces a new sampling strategy and shows that superior performance can be obtained for a range of sampling based robotic motion planners, used in scenarios with low task variance, as found in many vision guided pick and place operations. The strategy uses kernel density estimation to identify regions with high probability of containing configurations being part of feasible solutions, and use the estimation to bias sampling. The kernel densities are initialized with a uniform distribution and are continuously updated, whenever paths are successfully planned and optimized. The system is thereby self-learning and improves performance over time. The sampler is tested on a variety of planners and against other sampling methods in two different scenarios containing robotic arms with 6 degrees of freedom and compared to a state-of-the-art optimization based planning algorithm. Tests show that the sampler learns fast and improves both time taken for solving problems and the quality of the resulting paths compared to other samplers.

This paper proposes joint torque servo control of an electro-hydrostatic actuator (EHA) for effective dynamic control of robot manipulators. In the control design, the fluid parameters are not considered to develop a simple low-level controller used for multi degree-of-freedom (DOF) robot manipulators. Instead, mechanical impedance, which represents the effect of joint velocity on joint torque, is reduced using a disturbance observer. The torque tracking performance of the modified system is then improved by a simple proportional-integral-derivative control. When evaluated on two 1-DOF EHAs, the proposed controller guaranteed robustness to external disturbances and variations in the fluid parameters, even though the fluid parameters are not considered.

Commercialization of desktop milling machines has made rapid Printed Circuit Board (PCB) fabrication accessible. Unfortunately, PCB design for embedded and robotic systems is still a tedious and time consuming activity. In this paper, we present a technique, Just In Time Printed Circuit Board (JITPCB) for designing PCB systems at speeds commensurate with the capability of desktop PCB milling machines. We propose designing boards by writing software circuit generators that wire together and lay out circuit components in a hierarchical and reusable fashion. We have developed a declarative design mechanism allowing users to specify desired input and output peripherals and well as application code. Given this input, our system produces a complete working circuit board design, along with necessary initialization and networking code. Our system is an open framework that allows users to create a set of highly reusable parametric hardware/software modules. We demonstrate our approach by showing some common robotics applications designed with JITPCB to show its utility and generality.

We propose an online iterative path optimisation method to enable a Baxter humanoid robot to assist human users to dress. The robot searches for the optimal personalised dressing path using vision and force sensor information: vision information is used to recognise the human pose and model the movement space of upper-body joints; force sensor information is used for the robot to detect external force resistance and to locally adjust its motion. We propose a new stochastic path optimisation method based on adaptive moment estimation. We first compare the proposed method with other path optimisation algorithms on synthetic data. Experimental results show that the performance of the method achieves the smallest error with fewer iterations and less computation time. We also evaluate real-world data by enabling the Baxter robot to assist real human users with their dressing.

In this study, we propose a novel iterative learning scheme, which generates time-series data comprising the impedance value for human-robot cooperative work, where the human operator moves the end-effector from an initial position to a goal position. The proposed learning scheme iteratively updates time-series data comprising the variable impedance value to minimize a task-oriented cost function. Two types of noise reduction technique are used in the proposed learning scheme, which reduce the noise in the time direction and in the trial direction. The validity of the proposed method was verified based on experiments.

State-of-the-art techniques proposed for 6D object pose recovery depend on occlusion-free point clouds to accurately register objects in 3D space. To reduce this dependency, we introduce a novel architecture called Iterative Hough Forest with Histogram of Control Points that is capable of estimating occluded and cluttered objects' 6D pose given a candidate 2D bounding box. Our Iterative Hough Forest is learnt using patches extracted only from the positive samples. These patches are represented with Histogram of Control Points (HoCP), a “scale-variant” implicit volumetric description, which we derive from recently introduced Implicit B-Splines (IBS). The rich discriminative information provided by this scale-variance is leveraged during inference, where the initial pose estimation of the object is iteratively refined based on more discriminative control points by using our Iterative Hough Forest. We conduct experiments on several test objects of a publicly available dataset to test our architecture and to compare with the state-of-the-art.

Tactile data and kinesthetic cues are two important sensing sources in robot object recognition and are complementary to each other. In this paper, we propose a novel algorithm named Iterative Closest Labeled Point (iCLAP) to recognize objects using both tactile and kinesthetic information. The iCLAP first assigns different local tactile features with distinct label numbers. The label numbers of the tactile features together with their associated 3D positions form a 4D point cloud of the object. In this manner, the two sensing modalities are merged to form a synthesized perception of the touched object. To recognize an object, the partial 4D point cloud obtained from a number of touches iteratively matches with all the reference cloud models to identify the best fit. An extensive evaluation study with 20 real objects shows that our proposed iCLAP approach outperforms those using either of the separate sensing modalities, with a substantial recognition rate improvement of up to 18%.

This paper presents a learned, place-dependent terrain-assessment classifier that improves over time. Whereas typical methods aim to assess all of the terrain in a given environment, we exploit the fact that many robotic navigation tasks are well-suited to visual-teach-and-repeat navigation where robot motion is restricted to previously driven paths. In such scenarios, we argue that general terrain assessment is not required, and we can instead solve the much easier problem of detecting changes along the path; we sacrifice the ability to generalize off the path in favour of improved performance on the path. Terrain along a pretaught path is compared to terrain seen at the same location during previous, human-supervised traverses, and any significant differences cause that location to be labelled as unsafe. By storing all of our previous experiences, we are able to continuously improve our estimates as we revisit the same locations multiple times. We tested our method on two datasets collected at the University of Toronto and show that we improve over existing place-independent (both learned and not) methods, enabling nearly full autonomy in challenging, varied terrain using only a stereo camera.

In this paper, we propose a novel inverse reinforcement learning algorithm with leveraged Gaussian processes that can learn from both positive and negative demonstrations. While most existing inverse reinforcement learning (IRL) methods suffer from the lack of information near low reward regions, the proposed method alleviates this issue by incorporating (negative) demonstrations of what not to do. To mathematically formulate negative demonstrations, we introduce a novel generative model which can generate both positive and negative demonstrations using a parameter, called proficiency. Moreover, since we represent a reward function using a leveraged Gaussian process which can model a nonlinear function, the proposed method can effectively estimate the structure of a nonlinear reward function.

This paper introduces a new method for automatic robotic needle steering in deformable tissues. The main contribution relies on the use of an inverse Finite Element (FE) simulation to control an articulated robot interacting with deformable structures. In this work we consider a flexible needle, embedded in the end effector of a 6 arm Mitsubishi RV1A robot, and its insertion into a silicone phantom. Given a trajectory on the rest configuration of the silicone phantom, our method provides in real-time the displacements of the articulated robot which guarantee the permanence of the needle within the predefined path, taking into account any undergoing deformation on both the needle and the trajectory itself. A forward simulation combines i) a kinematic model of the robot, ii) FE models of the needle and phantom gel iii) an interaction model allowing the simulation of friction and puncture force. A Newton-type method is then used to provide the displacement of the robot to minimize the distance between the needle's tip and the desired trajectory. We validate our approach with a simulation in which a virtual robot can successfully perform the insertion while both the needle and the trajectory undergo significant deformations.

In this paper, we present a novel kinematic framework using hybrid twists, that has the potential to improve the reliability of estimated human shoulder kinematics. This is important as the functional aspects of the human shoulder are evaluated using the information embedded in thoracohumeral kinematics. We successfully demonstrate in our results, that our approach is invariant of the body-fixed coordinate definition, is singularity free and has high repeatability; thus resulting in a flexible user-specific kinematic tracking not restricted to bony landmarks.

Commercialized upper limb prostheses do not match the expectations of amputated people, especially transhumeral amputees. Most of them report a lack of functionality, mostly explained by a counter-intuitive control strategy. This paper presents the first implementation of an automatic prosthesis control approach based on natural coordinations between upper limb joints and IMU-based humeral orientation measurement. Two healthy individuals were able to use the prosthetic forearm attached to their upper arm to point at targets in a 3D workspace with a reasonable error. The results demonstrate the potential applications of automatizing the motion of some joints along the upper limb, in the same way as human upper limbs are controlled.

In this paper, we introduce a novel robot programming paradigm. It focuses on reducing the required expertise in robotics to a level that allows shop floor workers to use robots in their application domain without the need of extensive training. Our approach is user-centric and can interpret underspecified robot tasks, enabling communication on an abstract level. Such high-level task descriptions make the system amenable for users that are experts in a particular domain, but have limited knowledge about robotics and are thus not able to specify low-level details and instructions. Semantic models for all involved entities, i.e., processes, workpieces, and workcells, enable automatic reasoning about underspecified tasks and missing pieces of information. We showcase and evaluate this methodology on two industrial use cases from the domains of assembly and woodworking, comparing it to state-of-the-art solutions provided by robot manufacturers.

As robots aspire for long-term autonomous operations in complex dynamic environments, the ability to reliably take mission-critical decisions in ambiguous situations becomes critical. This motivates the need to build systems that have situational awareness to assess how quali ed they are at that moment to make a decision. We call this self-evaluating capability as introspection. In this paper, we take a small step in this direction and propose a generic framework for introspective behavior in perception systems. Our goal is to learn a model to reliably predict failures in a given system, with respect to a task, directly from input sensor data. We present this in the context of vision-based autonomous MAV flight in outdoor natural environments, and show that it effectively handles uncertain situations.