IROS 2016 Papers

This paper introduces GUMS, a complete projection model for omnidirectional stereo vision systems. GUMS is based on the existing generalized unified model (GUM), which we extend in order to satisfy a tight relationship among a pair of omnidirectional views for fixed baseline sensors. We exemplify the proposed model's calibration via a single-camera coaxial omnistereo system in a joint bundle-adjusted fashion. We compare our coupled method against the naive approach where the calibration of intrinsic parameters is first performed individually for each omnidirectional view using existing monocular implementations, to then solve for the extrinsic parameters as an additional step that has no effect on the intrinsic model solutions initially computed. We validate GUMS and its calibration effectiveness using both real and synthetic systems against ground-truth data. Our calibration method proves successful for correcting the unavoidable misalignment present in vertically-configured catadioptric rigs. We also generate 3D point clouds employing the calibrated GUMS systems in order to demonstrate the qualitative outcome of our contribution.

We present GRAVITY-ASSIST, a series elastic active body weight support and inertia compensation system for use in robot assisted gait rehabilitation. The device consists of a single degree of freedom series elastic actuator that connects to the trunk of a patient. The series elastic system is novel in that, it can provide the desired level of dynamic unloading such that the patient experiences only a percentage of his/her weight and inertia. Inertia compensation is important, since the inertial forces can cause significant deviations from the desired unloading force, specially at low support forces and fast walking speeds. Furthermore, this feature enables the inertia of the harness and force sensing unit attached to the patient to be compensated for, making sure that the device does not interfere with the natural gait cycle. We present a functional prototype of the device, its characterization and experimental verification of the approach.

The extreme variability in the appearance of a place across the four seasons of the year is one of the most challenging problems in life-long visual topological localization for mobile robotic systems and intelligent vehicles. Traditional solutions to this problem are based on the description of images using hand-crafted features, which have been shown to offer moderate invariance against seasonal changes. In this paper, we present a new proposal focused on automatically learned descriptors, which are processed by means of a technique recently popularized in the computer vision community: Convolutional Neural Networks (CNNs). The novelty of our approach relies on fusing the image information from multiple convolutional layers at several levels and granularities. In addition, we compress the redundant data of CNN features into a tractable number of bits for efficient and robust place recognition. The final descriptor is reduced by applying simple compression and binarization techniques for fast matching using the Hamming distance. An exhaustive experimental evaluation confirms the improved performance of our proposal (CNN-VTL) with respect to state-of-the-art methods over varied long-term datasets recorded across seasons.

This paper presents a novel structured knowledge representation called the functional object-oriented network (FOON) to model the connectivity of the functional-related objects and their motions in manipulation tasks. The graphical model FOON is learned by observing object state change and human manipulations with the objects. Using a well-trained FOON, robots can decipher a task goal, seek the correct objects at the desired states on which to operate, and generate a sequence of proper manipulation motions. The paper describes FOON's structure and an approach to form a universal FOON with extracted knowledge from online instructional videos. A graph retrieval approach is presented to generate manipulation motion sequences from the FOON to achieve a desired goal, demonstrating the flexibility of FOON in creating a novel and adaptive means of solving a problem using knowledge gathered from multiple sources. The results are demonstrated in a simulated environment to illustrate the motion sequences generated from the FOON to carry out the desired tasks.

The Jerboa, a tailed bipedal robot with two hip-actuated, passive-compliant legs and a doubly actuated tail, has been shown both formally and empirically to exhibit a variety of stable hopping and running gaits in the sagittal plane. In this paper we take the first steps toward operating Jerboa as a fully spatial machine by addressing the predominant mode of destabilization away from the sagittal plane: body roll. We develop a provably stable controller for underactuated aerial stabilization of the coupled body roll and tail angles, that uses just the tail torques. We show that this controller is successful at reliably reorienting the Jerboa body in roughly 150 ms of freefall from a large set of initial conditions. This controller also enables (and appears intuitively to be crucial for) sustained empirically stable hopping in the frontal plane by virtue of its substantial robustness against destabilizing perturbations and calibration errors. The controller as well as the analysis methods developed here are applicable to any robotic platform with a similar doubly-actuated spherical tail joint.

This work focuses on how to compute trajectories for an autonomous wheelchair based on indoor GIS maps, in particular on IndoorGML maps, which set the standard in this context. Good wheelchair trajectories are safe and comfortable for the user and the people sharing the space with him, turn gently, are high legible, and smooth (at least G2 continuos). We derive a navigation graph from a given IndoorGML map. We define and solve an optimization problem to find the desired path: given a succession of cells to traverse, the path corresponds to the best composite Bézier trajectory for the wheelchair. We discuss a related multi-objective path planning problem. Experimental results and an implementation on real robots show the planner performance.

This paper presents a robotic 3D printer: a robot system that can assemble toy brick sculptures from their 3D CAD models. In this system, a 3D CAD model is automatically converted to a block model consisting of primitive toy blocks. Then an assembly plan of the block model is automatically generated, if feasible. According to the plan, an industrial robot assembles a brick sculpture layer by layer from bottom to top. We demonstrate successful assembly of several brick sculptures.

Despite an intensive research on flapping flight and flapping wing MAVs in recent years, there are still no accurate models of flapping flight dynamics. This is partly due to lack of free flight data, in particular during manoeuvres. In this work, we present, for the first time, a comparison of free flight forces estimated using solely an on-board IMU with wind tunnel measurements. The IMU based estimation brings higher sampling rates and even lower variation among individual wingbeats, compared to what has been achieved with an external motion tracking system in the past. A good match was found in comparison to wind tunnel measurements; the slight differences observed are attributed to clamping effects. Further insight was gained from the on-board rpm sensor, which showed motor speed variation of ± 15% due to load variation over a wingbeat cycle. The IMU based force estimation represents an attractive solution for future studies of flapping wing MAVs as, unlike wind tunnel measurements, it allows force estimation at high temporal resolutions also during manoeuvres.

We describe a new decentralized algorithm for multi-robot systems to form arbitrary repeated lattice patterns. Prior work showed how to represent a desired pattern using a directed graph in which each edge is labeled with a rigid body transformation, and proposed an algorithm that accepts this graph as input and computes destinations for each robot using only local information. In this paper, we improve upon that result by describing a new algorithm, substantially different both in message passing procedure and in movement strategy, to resolve several limitations of the existing algorithm. We prove that, by executing this algorithm, the robots will form the desired lattice pattern in a bounded amount of time. We further show that, if the robots' communication graph is connected at the start of the algorithm, it will remain connected throughout the algorithm's execution. Using a simulation, we demonstrate that this algorithm works correctly for systems with dozens of autonomous robots to form various lattice patterns. Moreover, the experiments show a significant improvement in solution quality for our new algorithm compared to the previous approach.

We study formation change for robot groups in known environments. We are given a team of robots partitioned into groups, where robots in the same group are interchangeable with each other. A formation specifies the locations occupied by each group. The objective is to find collision-free paths that move all robots from a given start formation to a given goal formation. Our algorithm TAPF* has the following features: (a) it incorporates kinematic constraints of robots in form of velocity limits; (b) it maintains a user-specified safety distance between robots; (c) it attempts to minimize the makespan; and (d) it runs efficiently for hundreds of robots and dozens of groups even in dense 3D environments with narrow corridors and other occlusions. We demonstrate the efficiency and effectiveness of TAPF* in simulation and on robots.

In this paper, we introduce an approach to efficient robot navigation through cluttered indoor environments. We propose to estimate local obstacle densities based on already detected objects and use them to predict traversal costs corresponding to potential obstacles in regions not yet observable by the robot's sensors. By taking into account the predicted costs for path planning, the robot is then able to navigate in a more foresighted manner and reduces the risk of getting stuck in cluttered regions. We thoroughly evaluated our approach in simulated and real-world experiments. As the experimental results demonstrate, our method enables the robot to efficiently navigate through environments containing cluttered regions and achieves significantly shorter completion times compared to a standard approach not using any prediction.

For physical human-robot interaction (pHRI), it has been an important issue to control the output force of actuators. The aim of this study was to apply a new control strategy, named model-inverse time delay control (MiTDC), to series elastic actuators (SEA) in a lower extremity exoskeleton, even in the presence of uncertainties from pHRI. The law for time delay control (TDC) is derived and implementation issues are discussed, including the design of the state observer and the selection of the nominal value of the control distribution coefficient. Additionally, a new concept, a new reference position using the inverse of model dynamics, is introduced to realize satisfactory tracking performance without delay. Experimental results showed that the suggested controller achieved satisfactory performance without accurate information on system parameters, requiring only a nominal value of the control distribution coefficient.

In recent researches relating to robotic fingers, a novel power transmission mechanism had been designed, and developed as Twist-drive Actuators or Twisted String Actuator systems (TSA). These actuation structures, state of the art, include compact designs, light-weight mechanical structures, inherent compliance, and variable gearing, resulting in the fabrication of anthropomorphic robotic hands. However, those studies mentioned position control of the joint, and focused on the control performance of the pulling forces of a twisted string. This paper, therefore, introduces a novel joint mechanism composed of an Antagonistically-twisted Round-belt Actuator (ARA), which is able to make rotating motion by means of contraction forces induced by twisting small-diameter elastic round-belts. A noteworthy point in this paper is that the ARA robot is extremely well-suited for contact force control, which is exerted on the tip of the single joint robot. First, we demonstrate the existence of hysteresis characteristics of static contraction forces induced by the round-belt twisting, in which three belts are simultaneously twisted at constant low-speed. In addition, there exists a stress relaxation phenomenon when strongly twisting the elastic round-belts. We reveal that such sorts of discontinuous and nonlinear properties do not influence control performance in either the contraction force on the belts or contact force on the tip of the robot. Finally, this paper clearly shows stable and accurate tracking performance of the contact force of the ARA robot. In these experiments, we newly develop a twin/triple-twisted round-belt structure for the agonist side actuator, thus enhancing the contact forces effectively.

This paper introduces a method of force control for a jumping biped robot to correctly control the ground reaction force during continuous jumping. Assuming dynamic motion, such as jumping and running, the attitude of the robot depends on the dynamics in response to the ground reaction force. Therefore, controlling the ground reaction force is necessary for stabilizing the robot's motion. However, control of the ground reaction force involves a fundamental problem: feedback control does not work properly against the impact force owing to limitations in the control bandwidth. This study introduces a method for stiffness ellipse control that utilizes three antagonist pairs of six pneumatic artificial muscles. When the tip of the robot makes contact with the ground, an external force is induced in the direction of lower stiffness. By utilizing this property of the stiffness ellipse, it is possible to control the ground reaction force by feedforward control, which is not dependent on the control bandwidth. Based on this idea, impact force control at landing and jumping force control at takeoff were proposed to correctly control the ground reaction force during the continuous jumping of the robot. The results of several experiments conducted convince us that the relationship between the ground reaction force and the stiffness ellipse is almost linear, and that the ground reaction force can be controlled with high reproducibility by adjusting the stiffness ellipse.

Proximity and distance estimation sensors are broadly used in robotic hands to enhance the quality of grasping during grasp planning, grasp correction and in-hand manipulation. This paper presents a fiber optical proximity sensor that is integrated with a tactile sensing fingertip of a robotic hand of a mobile robot. The distance estimation of proximity sensors are typically influenced by the reflective properties of an object, such as color or surface roughness. With the approach proposed in this paper, the accuracy of the proximity sensor is enhanced using the information collected by the vision system of the robot. A camera is employed to obtain RGB values of the object to be grasped. Further on, the data obtained from the camera is used to obtain the correct calibration for the proximity sensor. Based on the experimental evidence, it is shown that our approach can be effectively used to reduce the distance estimation error.

Self-folding provides an efficient way of creating complex 3D geometries from 2D composites. However, the precision of self-folding structures is often limited by the use of open-loop folding mechanisms. In this paper we demonstrate feedback-controlled self-folding using a shape memory polymer and optical sensors to accurately control folding angles. We present a method of quickly and inexpensively fabricating large collectives of self-folding autonomous robots that can be transported in flat configurations prior to autonomous deployment at target destinations. To demonstrate this, we build a collective of robots that is manufactured in one continuous laminar composite. Individual robots in the collective detach from one other, self-fold into pre-programmed configurations and navigate by phototaxis. This method could be applied to a broad range of applications where logistics necessitate compact transport and where external manipulation is difficult or expensive, such as in space applications or delivering search-and-rescue robots in cluttered environments.

The paper describes a method of fault-tolerant adaptive gait generation based on CPG controller with interlimb coordination for multi-limbed working robot. Multi-limbed robot is expected to work in dangerous, complicated narrow spaces, where human workers cannot reach. However, if one of the limbs is broken in the dangerous/narrow working space, human operators cannot go to the space to repair it. Even in these situations, the proposed method generates a new gait adaptively using the remaining usable limbs depending on the current situation of the multi-limbed robot. This fault-tolerant ability is effective for multi-limbed robots working in dangerous/narrow space. The method is implemented to a model of a multi-limbed robot “ASTERISK” in a dynamical simulator (ODE). Experimental results show effectiveness of the proposed method.

Obstacle Detection is a central problem for any robotic system, and critical for autonomous systems that travel at high speeds in unpredictable environment. This is often achieved through scene depth estimation, by various means. When fast motion is considered, the detection range must be longer enough to allow for safe avoidance and path planning. Current solutions often make assumption on the motion of the vehicle that limit their applicability, or work at very limited ranges due to intrinsic constraints. We propose a novel appearance-based Object Detection system that is able to detect obstacles at very long range and at a very high speed (~ 300Hz), without making assumptions on the type of motion. We achieve these results using a Deep Neural Network approach trained on real and synthetic images and trading some depth accuracy for fast, robust and consistent operation. We show how photo-realistic synthetic images are able to solve the problem of training set dimension and variety typical of machine learning approaches, and how our system is robust to massive blurring of test images.

Object segmentation from 3D range data is an important topic in mobile robotics. A robot navigating in a dynamic environment needs to be aware of objects that might change or move. A segmentation of the laser scans into individual objects is typically the first processing step before a further analysis is performed. In this paper, we present a fast method that segments 3D range data into different objects, runs online, and has small computational demands. Our approach avoids the explicit computation of the 3D point cloud and performs all computations directly on a 2D range image, which enables a fast segmentation for each scan. A further relevant aspect of our method is that we can segment objects even if the 3D data is sparse. This is important for scanners such as the new Velodyne Puck. We implemented our approach in C++ and ROS and thoroughly tested it using different 3D scanners. Our method can operate at over 100 Hz for the 64-beam Velodyne scanner on a single core of a mobile CPU while producing high quality segmentation results. In addition to this, we make the source code for the approach available.