IROS 2015 论文列表

Any system that realizes the function of autonomous driving has to perform the task of constructing a continuous trajectory that can be followed by the autonomous vehicle. Many planning approaches may benefit from an efficient algorithm that is able to construct a continuous trajectory which satisfies the initial and target poses of the vehicle as well as the vehicle actuator states at the boundaries. We present an approach to planning motion trajectories that solves this problem. Our method uses a steering profile which is inspired by human driving behavior. Careful parametrization of the steering profile allows designing of an efficient numerical solver to compute the values of the profile parameters for a wide range of possible vehicle target states. In addition, the comprehensible interpretation of the profile parameters may be exploited by planners that rely on trajectory sampling with varying boundary conditions at the trajectory end point.

Environment recognition is an effective way for a mobile robot to move across rough terrain. In particular, this makes it possible to prevent a tiny mobile robot from getting stuck or turning over. Several studies have been conducted on environment recognition using a laser range finder or camera. However, almost all of these studies focused on obstacle detection or shape recognition, which cannot be used to recognize the surface condition such as slipperiness. The purpose of this work is to design a model for estimating the surface condition using a tiny mobile robot. We set slipperiness as one of the parameters for recognizing the surface condition, which is already used by terramechanics, along with two additional parameters, the hardness and unevenness. We find that a robot can roughly estimate the ground hardness by measuring the current peak of a motor and the unevenness from measuring the robot posture. By recognizing the surface condition, the robot can change the parameters of the controlling motor based on the ground characteristics. This new method for recognizing the surface condition is significant, not only because it fills gaps in the previous research, but also because it does not require any special sensors such as a laser range finder and does not consume a large quantity of energy. Therefore, it achieves a core objective of our environmental monitoring system using multiple mobile robot.

Recent studies showed that Continuously Variable Transmissions (CVT) and Infinitely Variable Transmissions (IVT) can considerably improve the locomotion efficiency in legged robot. A CVT is a transmission whose ratio can be continuously varied and an IVT is a transmission whose ratio can be continuously varied from positive to negative values. However, efficient use of such transmissions in walking applications requires changing the transmission ratio at a minimal energy cost, even at rest, i.e. when the input shaft is not rotating. This contribution proposes a novel CVT and IVT principle which can achieve such ratio variations at rest. The presented CVT is a modified planetary gear, whose planets are conical and mounted on inclined shafts, and whose ring is made of contiguous diabolo-shaped rollers. This configuration enables the control of the transmission ratio by adjusting the point of contact between the cones and rollers that comprise the ring. A traditional planetary gear system can be added to the CVT to form an IVT.

We present a method for segmenting a set of unstructured demonstration trajectories to discover reusable skills using inverse reinforcement learning (IRL). Each skill is characterised by a latent reward function which the demonstrator is assumed to be optimizing. The skill boundaries and the number of skills making up each demonstration are unknown. We use a Bayesian nonparametric approach to propose skill segmentations and maximum entropy inverse reinforcement learning to infer reward functions from the segments. This method produces a set of Markov Decision Processes (MDPs) that best describe the input trajectories. We evaluate this approach in a car driving domain and a simulated quadcopter obstacle course, showing that it is able to recover demonstrated skills more effectively than existing methods.

We consider the problem of state-observation and control for a bi-tethered aerial system composed by a physical chain of two underactuated aerial robots, also called UAVs. The controlled outputs are the Cartesian position of the last robot and the internal forces along the links. We aim at a minimal use of sensors in order to retrieve the full state. For this goal we propose an output transformation method whose applicability implies the system observability. When this is the case we prove that it is possible to design a nonlinear state estimator based on the high gain- and Luenberger- observers that is able to retrieve the state from any dynamic condition. We also demonstrate how this estimator can be employed with a nonlinear controller for the Cartesian position and the link stresses while ensuring the stability in closed-loop. We show the validity of the method for sensorial configurations composed only by two accelerometers (no gyros) and just two encoders, or two accelerometers (no gyros) and just two inclinometers. A realistic simulative validation concludes the paper.

This paper addresses modeling issues for a class of nano-robotic systems using piezoelectric stick-slip actuators. The work focuses on the friction force modeling to describe the dynamics of a stick-slip actuator in a wide operating range needed in nano-robotics. Based on the theory of the single state elasto-plastic model and on an experimental analysis, necessary conditions on presiding modeling are highlighted. The conditions allow describing the dynamics of stick-slip type actuators for both scanning mode and stepping mode in the time and the frequency domains and for backward and forward directions of the motion. The proposed dynamic model opens new perspective for closed loop control of nano-robotic system.

This paper proposes a simple solution for stabilization of a nano-hexacopter carrying a manipulator arm in order to increase the type of missions achievable by these types of systems. The manipulator arm is attached to the lower part of the hexacopter. The motion of the arm induces a change of the center of mass of the whole body, which induces torques which can produce the loss of stability. The present work deals with the stabilization of the whole system - that is hexacopter and arm - by means of a set of nonlinear control laws. First, an attitude control, stabilizes the hexacopter to a desired attitude taking into account the movement of the arm. Then, a suitable virtual control and the translational dynamics allow the formulation of a nonlinear controller, which drives the aerial vehicle to a desired position. Both controls consist in saturation functions. Experimental results validate the proposed control strategy and compares the results when the motion of the arm is taken into account or not.

Small insects can use vision to successfully land on arbitrary surfaces. For miniature robots with a size of a few centimeters, it is challenging to achieve similar capabilities due to their small sizes. To address this problem, we present a new vision based landing algorithm using a non-vector space control method, which formulates the system dynamics in the space of sets—different from the traditional vector space with a linear structure. Using a stabilization controller in the non-vector space, we can control the landing posture of a small robot with an active tail. Preliminary experimental results demonstrate the effectiveness of the proposed approach. The research presented in this paper, together with the idea of compressive feedback—when only a compressed image is used for feedback—can potentially achieve real time onboard vision based control for miniature robots.

In contact micromanipulation, the adhesive forces between manipulators and microobjects decrease the chances of achieving successful releases at the desired positions. We study a non-contact micromanipulation technique of microbeads (300 µm in average diameter) using clusters of paramagnetic microparticles (100 µm in average diameter). This non-contact micromanipulation is done using the hydrodynamic forces instead of the interaction forces in contact manipulation, and hence eliminates the adhesive forces that decrease the chances of achieving successful releases. Motion of the cluster of microparticles results in a pressure gradient (within the vicinity of the microbead in a fluid) that derives and steers the microbeads without contact. The microparticles are moved under the influence of controlled magnetic field gradient to push or pull the microbeads towards reference positions. We achieve non-contact manipulation via pushing and pulling at average speeds of 219 µm/s and 258 µm/s for the microbead, respectively (using cluster of 10 microparticles). The noncontact pushing and pulling localize the microbeads within the vicinity of reference positions with average steady-state errors of 177 µm and 100 µm, respectively. Moreover, we experimentally demonstrate non-contact microassembly of 3 microbeads into an L-shape at a task completion time of 25 seconds.

This paper presents the BirthSIM model, a haptic interface of a childbirth simulator. The aim of this haptic interface is to improve haptic rendering and increase the biofidelity of the movements and forces which occur during the fetal descent. Thus the model of this 2 degrees of freedom pneumatic actuated interface is presented in this article, in order to synthesize a control law based on the backstepping method. The control law obtained allows targeted trajectories of positions and pneumatic stiffness to be tracked but also allows the equivalent stiffness of the end effector in Cartesian coordinates to be adapted in real time. The difference between pneumatic stiffness and closed loop stiffness is explained. Finally, some simulation results are presented to validate the control law behavior.

Braitenberg vehicles are simple models of the motion of animal towards, or away from, a stimulus. They have been used to implement target reaching and avoidance behaviours in robotics, based, among others, on sound, light, distance, and pressure sensors. When the sensors are accurate enough - when they provide a high signal to noise ratio - variations on the readings can be neglected. Their behaviour, in these applications, can be explained using a non-linear deterministic dynamical system. For noisy sensors, or when the physical interaction between the robot and the measured variable is complex, a deterministic model is not good enough. Some examples include robots with cheap sensors, sensor prototypes, or settings where the interaction with the environment changes the measurements, like odour tracking - as the motion of the robot creates turbulences in the air. This paper presents the first analysis of the behaviour of Braitenberg vehicle 3a with noisy sensors. The mathematical equations of the evolution of the vehicle state probability are derived under white noise assumptions, and a bound for the vehicle state uncertainty is obtained assuming Gaussian probability distributions of the pose. These equations relate the morphological parameters of the vehicle, the noise variance and the function connecting the sensors to the actuators. The non-linear controller simulations illustrate the local validity of the model, and global simulations show how the vehicles converge.

In this study, we propose a neural network based model for learning a robot's drawing sequences in an unsupervised manner. We focus on the ability to learn visual-motor relationships, which can work as a reusable memory in association of drawing motion from a picture image. Assuming that a humanoid robot can draw a shape on a pen tablet, the proposed model learns drawing sequences, which comprises drawing motion and drawn picture image frames. To learn raw pixel data without any given specific features, we utilized a deep neural network for compressing large dimensional picture images and a continuous time recurrent neural network for integration of motion and picture images. To confirm the ability of the proposed model, we performed an experiment for learning 15 sequences comprising three types of shapes. The model successfully learns all the sequences and can associate a drawing motion from a not trained picture image and a trained picture with similar success. We also show that the proposed model self-organizes its behavior according to types shapes.

In this paper, we present a holistic system for automating the motion of a rolling microrobot for protein crystal harvesting. The RodBot, which was introduced in previous work, is able to perform noncontact manipulation of microscopic objects such as fragile crystals by trapping them in an induced vortex fluid flow. Here, we are concerned with navigating the RodBot autonomously in a liquid environment containing obstacles such as crystals. A literature review shows existing approaches to untethered microrobot control are limited and cluttered environments are often not considered. We demonstrate real-time tracking of the RodBot and surrounding obstacles, kinematic obstacle-free path planning, and nonholonomic path following. The system was evaluated in qualitative and quantitative experiments, shows satisfactory performance, and presents itself as a first step towards fully automated crystal harvesting.

Navigation in complex and unknown environments is a major challenge for blind people. The most popular, conventional navigation aids such as white canes and guide dogs, however, provide limited assistance in such settings as they are constrained to interpret the local environment only. At the same time, they can hardly be combined with a walker required by elderly people with walking disabilities. Technologies developed in the field of robotics have the potential to assist blind people in complex navigation tasks as they can provide information about obstacles and reason on both global and local environment models. The contribution of this paper is a smart walker that enables blind users to safely navigate. It includes an innovative vibro-tactile user interface and a controller that takes into account human characteristics based on a user study. The walker has been designed to deal with the fact that humans can only sense and interpret a limited number of commands and have a delayed response. Our experiments validate our claim that the technique outlined in this paper guides a user to the desired goal in less time and with shorter traveled distance compared to a standard robotic controller.

In this paper we present a novel on-line method to recursively align point clouds. By considering each point together with the local features of the surface (normal and curvature), our method takes advantage of the 3D structure around the points for the determination of the data association between two clouds. The algorithm relies on a least squares formulation of the alignment problem, that minimizes an error metric depending on these surface characteristics. We named the approach Normal Iterative Closest Point (NICP in short). Extensive experiments on publicly available benchmark data show that NICP outperforms other state-of-the-art approaches.

The problem of active perception with an autonomous robot is studied in this paper. It is proposed that the exploratory behavior of the robot be controlled using mutual information (MI) surfaces between the current map and a one-step look ahead measurements. MI surfaces highlight informative areas for exploration. A novel method for computing these surfaces is described. An approach that exploits structural dependencies of the environment and handles sparse sensor measurements to build a continuous model of the environment, that can then be used to generate MI surfaces is also proposed. A gradient field of occupancy probability distribution is regressed from sensor data as a Gaussian Process and provide frontier boundaries for further exploration. The continuous global frontier surface completely describes unexplored regions and, inherently, provides an automatic termination criterion for a desired sensitivity. The results from publicly available datasets confirm an average improvement of the proposed methodology over comparable standard and state-of-the-art exploratory methods available in the literature by more than 20% and 13% in travel distance and map entropy reduction rate, respectively.

This paper presents a centralized algorithm for multi-scale observation of multiple moving targets using a team of Micro Aerial Vehicles (MAVs). The proposed algorithm is appropriate when MAVs can observe targets at different elevations with the objective of jointly maximizing duration and resolution of observation for each target. The MAVs share the workload using a greedy assignment of locations and targets to MAVs. The proposed algorithm uses a quad-tree data structure to model the movement decisions of MAVs as well as the variable qualities (resolutions) of observations. We consider cases where there is uncertainty in the target observations (i.e., measurement noise), the number of targets is larger than that of the MAVs and the combined field of views (FOVs) of the sensors cannot cover the whole search region. Simulation results confirm the effectiveness of the proposed algorithm.

The number of artificial space debris around the earth has been increasing, and it is necessary to remove this debris to prevent collision with satellites in operation. One of the most practical methods which remove large debris, such as non-functioning satellite, is attaching a propulsion system for reentry. However, it exists difficulty to attach the system because non-cooperative satellites are expected to have no fixture specialized for grasping purpose. Furthermore, the dynamic parameters of the target need to be determined to realize stabilization when they are unknown. Accordingly, multipoint contact is introduced for stabilization of push-only control and estimate parameters. This paper describes the estimation method of the dynamic parameters using an unscented Kalman filter based on multipoint contact information and attitude stabilization using push-only (without pull operation) control based on feedback linearization and receding horizon with input constraints.

Simultaneous mastering of multiple tasks during motion generation is challenging. Traditional null-space based approaches for redundant robots implement a strict, hierarchical prioritization for tracking multiple objectives. In consequence, these schemes are not suited to impose smooth priorities or changing them during motion execution. A recently developed mixture of controller approach superpose torques from several control modules for motion generation and thereby enables to flexibly impose priorities for pursuing different goals in parallel. The main contribution of this paper is the development of a framework which allows for automatic derivation of suitable mixture coefficients which represent priorities. The functionality of the optimization framework is demonstrated for a virtual 3 DOF pendulum and the humanoid robot COMAN.