IROS 2018 论文列表

This paper presents a face portrait with cartoon stylization painting and associated algorithms with the visual feedback system to paint like a human cartoonist. The robot cartoonist creates the artwork in two stages-cartoon style transformation and robot artist for colorful painting. In the cartoon style transformation stage, it transfers human portrait photos to cartoon style by face detection and alignment, which can effectively decompose the face into individual components then replace by cartoon facial components. In the second stage, the robot uses an eye-in-hand system to obtain five basic colors (cyan, magenta, yellow, white and black) to automatically mix a variety of colors automatically. For painting strategy, we start with the outline of the face, which we use non-photorealistic rendering (NPR) to generate hand-painted strokes. After that, the robot artist will implement painting the facial features. We also demonstrate the success of this proposed research.

This paper presents a new model to make robots capable of approaching and engaging people with a human-like behavior, while they are walking in a side-by-side formation with a person. This method extends our previous work [1], which allows the robot to adapt its navigation behaviour according to the person being accompanied and the dynamic environment. In the current work, the robot is able to predict the best encounter point between the human-robot group and the approached person. Then, in the encounter point the robot modifies its position to achieve an engagement with both people. The encounter point is computed using a gradient descent method that takes into account all people predictions. Moreover, we make use of the Extended Social Force Model (ESFM), and it is modified to include the dynamic goal. The method has been validated over several situations and in real-life experiments, in addition, a user study has been realized to reveal the social acceptability of the robot in this task.

In this paper, we propose a novel robocentric formulation of visual-inertial navigation systems (VINS)within a multi-state constraint Kalman filter (MSCKF)framework and develop an efficient, lightweight, robocentric visual-inertial odometry (R-VIO)algorithm for consistent localization in challenging environments using only monocular vision. The key idea of the proposed approach is to deliberately reformulate the 3D VINS with respect to a moving local frame (i.e., robocentric), rather than a fixed global frame of reference as in the standard world-centric VINS, and instead utilize high-accuracy relative motion estimates for global pose update. As an immediate advantage of using this robocentric formulation, the proposed R-VIO can start from an arbitrary pose, without the need to align its orientation with the global gravity vector. More importantly, we analytically show that the proposed robocentric EKF-based VINS does not undergo the observability mismatch issue as in the standard world-centric frameworks which was identified as the main cause of inconsistency of estimation. The proposed R-VIO is extensively tested through both Monte Carlo simulations and real-world experiments using different sensor platforms in different environments and shown to achieve competitive performance with the state-of-the-art VINS algorithms in terms of consistency, accuracy and efficiency.

Due to the risks of muscle, bone and neural damage in spinal surgical procedures that require pedicle screw fixation, technological improvements have appeared to help surgeons perform the procedures with higher accuracy. Systems based on optical tracking navigation impose a stringent limitation in the workflow of surgeons since a clear line of sight has to be kept between the cameras and the tracked elements. Other solutions are based on mounting a miniature robot on the spine of the patient, which is very invasive and entails some risks. For these reasons, a novel robotic assisted surgery system capable to guide surgical instruments with minimal deviations compensating patient motion is being developed. This paper presents the system and the electromechanical tracking device used to sense patient motion.

In this paper, we propose a risk-based coordination method for the Multi-Robot Task Allocation (MRTA) problem in human-populated environments. We introduce risk-based bids that incorporate human trajectory prediction uncertainties and furthermore, social costs in their formulation. We demonstrate the effectiveness of including a predictive component in the risk formulation despite the lack of accurate position estimation for humans through an extensive suite of experiments. This is done by means of testing different levels of prediction error for known human trajectories and in a separate approach, using a Kalman filter for human trajectory estimation. Furthermore, we propose different risk formulations and evaluate their performance in a high-fidelity simulator. Additionally, a comparative study targeting human-agnostic planning at both navigation and planning levels, human-aware navigation and planning based on deterministic costs, and risk-based human-aware planning with no individual human-aware navigation has been conducted. Results confirm that risk-based bids lead to more socially acceptable team plans that reduce the need for the lower level individual human-aware navigation to be activated. Risk-based plans accounting for social costs prevent difficult social situations that can lead to less effective human-aware navigation, such as traversing narrow passages occupied by humans.

The sequential online tuning for controller gains is required for the continuous action of the riding into parallel two-wheeled scooter and the speed governing after riding by humanoid robot. The implemented controllers are different between the riding and the speed governing, and these tuning strategies are also different. In particular, the riding requires the immediate tuning in the short riding phase and the speed governing requires the accurate tuning to regulate the speed of humanoid robot. To the above requirements, this paper proposes the Sequential Online Learning Control (SOLC)method composed of the cascade connection of SGD-based open-loop Learning Control (SLC)and Mini-batch-based closed-loop Learning Control (MLC). SLC contributes the damping gain online tuning for the foot torque control during execution of riding, and MLC contributes the PID gains online tuning for the speed governing control. Finally, we show the validity of SOLC through the sequential experiment of riding and speed governing for parallel two-wheeled scooter by life-sized humanoid robot HRP2-JSK.

In this work, the authors report the investigation of liposomes' stability as a drug delivery vehicle, using the method of resistive pulse method. The main objects of interest are the 50nm diameter liposomes, while the 100nm diameter liposomes are widely used for its stability. However, certain drug delivery scenarios arise, like tighter cancerous tissue arrangement and different circulation time requirement, which dictates the necessity of sub-100nm diameter vesicles. The size measurements upon freshly fabricated liposomes are performed frequently on increasing time interval. The results exhibit a trend of size increasing, suggesting the existence of liposome fusion. The possible models of fusion are proposed and discussed. This work demonstrates the localized organic nanoparticle measurement with fine dual 3D manipulator positioning, which paves the way for the possible cellular in-vivo measurement within an environmental SEM chamber.

Applications of safety, security, and rescue in robotics, such as multi-robot target tracking, involve the execution of information acquisition tasks by teams of mobile robots. However, in failure-prone or adversarial environments, robots get attacked, their communication channels get jammed, and their sensors may fail, resulting in the withdrawal of robots from the collective task, and consequently the inability of the remaining active robots to coordinate with each other. As a result, traditional design paradigms become insufficient and, in contrast, resilient designs against system-wide failures and attacks become important. In general, resilient design problems are hard, and even though they often involve objective functions that are monotone or submodular, scalable approximation algorithms for their solution have been hitherto unknown. In this paper, we provide the first algorithm, enabling the following capabilities: minimal communication, i.e., the algorithm is executed by the robots based only on minimal communication between them; system-wide resiliency, i.e., the algorithm is valid for any number of denial-of-service attacks and failures; and provable approximation performance, i.e., the algorithm ensures for all monotone (and not necessarily submodular) objective functions a solution that is finitely close to the optimal. We quantify our algorithms approximation performance using a notion of curvature for monotone set functions. We support our theoretical analyses with simulated and real-world experiments, by considering an active information gathering scenario, namely, multi-robot target tracking.

Humans play sports to improve their athletic ability. The robot, especially humanoid robot, is also able to improve its athletic performances, such as reaction speed and balancing, through robot sports. Therefore, robots have been developed through performing various robot sports events such as robot soccer, robot marathon, robot fight and so on. In this reason, The Ski Robot Challenge was held in South Korea in commemoration of the PyeongChang 2018 Winter Olympic Games. The event was an Alpine slalom skiing competition in the almost same rules to human's but on a relatively short course (80m). To participate in this ski tournament, the skiing robot DIANA has been developed. In this video, the skiing robot technologies were introduced. At first, she must be able to recognize the flags. The deep learning method was used to recognize them. Secondly, she had a motion pattern to perform the carving turn, the most difficult and fastest skiing technique. In order to improve the stability, she compensated her motion to follow reference COP, based on the measured F/T sensor data. In addition, IMU sensor was used to remove instantaneous disturbance. Using these methods, the humanoid robot, DIANA, that can perform the carved turn on a realworld slope was successfully developed.

The research on robot tracking has focused on the problem of information fusion from redundant parametric estimations, though the aspect of choosing an adaptive fusion policy, that is computationally efficient, and is able to reduce the impact of un-modeled noise, are still open issues. The objective of this work is to study the problem of underwater robot localization. For this, we have considered a task relying on inertial and geophysical sensory. We propose an heuristic model that performs adaptable fusion of information based on the principle of contextually anticipating the localization signal within an ordered neighborhood, such that a set of nodes properties is related to the task context, and the confidence on individual estimates is evaluated before fusing information. The results obtained show that our model outperforms the Kalman filter and the Augmented Monte Carlo Localization algorithms in the task.

This paper addresses a new attitude determination problem for formations. It considers a three-vehicle formation with relative and inertial measurements from sensors, where Constraints limit the relative measurements, which are not available between two of the vehicles, also known as deputies. The other vehicle is called the chief and does not have any limitation. Furthermore, each of the vehicles has an independent inertial measurement, whose references are known. The goal is to determine all attitude relations, both inertial and relative. The solution for this problem is divided into different stages. First, the relative attitude between the chief and the deputies is assessed, which results in two candidates for each of these relations. Then, each candidate yields a candidate for the inertial attitude of the chief. Next, comparing the four inertial candidates gives the solution for their respective relations and consequently for the relative relations as well. The remaining relations derive directly from those already known. The paper also provides some early insights about degeneracies, possible particular cases of the solution, and the effect of sensor noise. Finally, the solution is validated with a simulation, whose results are similar to attitude determination problems in constrained formations.

It is well known that reinforcement learning (RL) can benefit from the use of a dynamic prediction model which is learned on data samples collected online from the process to be controlled. Most RL algorithms are formulated in the state-space domain and use state-space models. However, learning state-space models is difficult, mainly because in the vast majority of problems the full state cannot be measured on the system or reconstructed from the measurements. To circumvent this limitation, we propose to use input-output models of the NARX (nonlinear autoregressive with exogenous input) type. Symbolic regression is employed to construct parsimonious models and the corresponding value functions. Thanks to this approach, we can learn accurate models and compute optimal policies even from small amounts of training data. We demonstrate the approach on two simulated examples, a hopping robot and a 1-DOF robot arm, and on a real inverted pendulum system. Results show that our proposed method can reliably determine a good control policy based on a symbolic input-output process model and value function.

State abstraction in a discrete reinforcement learning setting clusters states sharing a similar optimal action to yield an easier to solve decision process. In this paper, we generalize the concept of state abstraction to continuous action reinforcement learning by defining an abstract state as a state cluster over which a near-optimal policy of simple shape exists. We propose a hierarchical reinforcement learning algorithm that is able to simultaneously find the state space clustering and the optimal sub-policies in each cluster. The main advantage of the proposed framework is to provide a straightforward way of regularizing reinforcement learning by controlling the behavioral complexity of the learned policy. We apply our algorithm on several benchmark tasks and a robot tactile manipulation task and show that we can match state-of-the-art deep reinforcement learning performance by combining a small number of linear policies.

We consider the problem of building accurate three dimensional (3D)reconstructions of orchard rows. This problem arises in many applications including yield mapping and measuring traits (e.g. trunk diameters)for phenotyping. While 3D reconstructions of side views can be obtained using standard methods, merging the two side-views is difficult due to the lack of overlap between the two partial reconstructions. We present a novel method that utilizes global features to constrain the solution. Specifically, we use information from the silhouettes and the ground plane for alignment. The method is evaluated using multiple simulated and real datasets. For additional information and demonstration of experimental results please see https://www.youtube.com/watch?v=6mGMF2gFv4M.

Increasingly, robots have more degrees of freedom (DOF), imposing a need for calculating more complex dynamics. As a result, better efficiency in carrying out dynamics computations is becoming more important. In this study, an efficient method for computing the joint space inertia matrix (JSIM) for high DOF serially linked robots is addressed. We call this method the Geometric Dynamics Algorithm for High number of robot Joints (GDAHJ). GDAHJ is non-symbolic, preserve simple formulation, and it is convenient for numerical implementation. This is achieved by simplifying the way to recursively derive the mass-matrix exploiting the unique property of each column of the JSIM and minimizing the number of operations with O(n2) complexity. Results compare favorably with existing methods, achieving better performance over state-of-the-art by Featherstone when applied for robots with more than 13 DOF.

Effective human-robot collaboration in shared control requires reasoning about the intentions of the human user. In this work, we present a mathematical formulation for human intent recognition during assistive teleoperation under shared autonomy. Our recursive Bayesian filtering approach models and fuses multiple non-verbal observations to probabilistically reason about the intended goal of the user. In addition to contextual observations, we model and incorporate the human agent's behavior as goal-directed actions with adjustable rationality to inform the underlying intent. We examine human inference on robot motion and furthermore validate our approach with a human subjects study that evaluates autonomy intent inference performance under a variety of goal scenarios and tasks, by novice subjects. Results show that our approach outperforms existing solutions and demonstrates that the probabilistic fusion of multiple observations improves intent inference and performance for shared-control operation.

We compare the ability of two different robot controllers for collective foraging to cope with noise in robot global positioning data and show how recruitment, in the form of broadcast messages near worksites, can make swarms more robust. Swarms of five e-puck robots are used in a semi-virtual environment, facilitated by the VICON positioning system. This setup allows us to control the amount of noise in the robot positioning data and to generate pseudo-random environments, while retaining important physical aspects of the experiment. The effect of inherent noise in the robot infra-red sensors, used for obstacle avoidance, is noted and the importance of modelling such noise in agent-based simulations is highlighted.