IROS 2018 论文列表

One of the main objectives of ubiquitous robots is to proactively provide context-aware intelligent services to assist humans in their professional or daily living activities. One of the main challenges is how to automatically obtain a consistent and correct description of human context such as location, activities, emotions, etc. In this paper, a new hybrid approach for reasoning on the context is proposed. This approach focuses on human activity recognition and consists of machine-learning algorithms, an expressive ontology representation, and a reasoning system. The latter allows detecting the inconsistencies that may appear during the machine learning phase. The proposed approach can also correct automatically these inconsistencies by considering the context of the ongoing activity. The obtained results on the Opportunity dataset demonstrate the feasibility of the proposed method to enhance the performance of human activity recognition.

Robotic teleoperation is a key technology for a wide variety of fields. Teleoperating a humanoid in particular is essential as it allows the user to act remotely on an interface designed especially for humans, e.g., in a space station, or operating tools and machinery in disaster scenarios. This paper presents a ‘task-relevant’ haptic interface for humanoid teleoperation, which bridges the gap between the task at hand and the balance of the robot. The operator is given command over the humanoid's hands and is informed through haptic cues about the impact of her/his potential actions on the robot’ stability. Moreover, a null-space autonomous controller acts in the operator's null-space to provide her/him with a wider workspace and help in the successful execution of the task. The architecture is designed to top an existing compliance controller for a torque-controlled humanoid robot. Experiments on the humanoid robot TORO are reported to demonstrate the feasibility and effectiveness of the approach.

This work proposes a humanoid robot center of mass (COM) estimation framework based on the compliant inverted pendulum model and the robust estimator. Humanoids' limited structural stiffness and relatively long legs result in undesired flexibility, and this undesired motion hinders the state estimation. The models used in previous studies were either not suitable for estimation or too simple to express these key characteristics of humanoid robots. Here, to enhance the estimation performance, the compliant inverted pendulum model, which is developed by attaching a spring and damper to the original pendulum, is adopted. The additional elements can represent the mechanical deformation and undesired flexibility. This model can reflect the important characteristics of the humanoid robot while taking advantage of the merits of the sing-mass model. In addition, a robust state estimator that was proposed in our previous work is adopted to compensate for an estimation error caused by a modeling error. Using these two factors, an improved COM kinematics estimates could be obtained.

Humanoids' abilities to navigate stairs and uneven terrain make them well-suited for disaster response efforts. However, humanoid navigation in such environments is currently limited by the capabilities of navigation planners. Such planners typically consider only footstep locations, but planning with palm contacts may be necessary to cross a gap, avoid an obstacle, or maintain balance. However, considering palm contacts greatly increases the branching factor of the search, leading to impractical planning times for large environments. In previous work we explored using library-based methods to address difficult navigation planning problems requiring palm contacts, but such methods are not efficient when navigating an easy-to-traverse part of the environment. To maximize planning efficiency, we would like to use discrete planners when an area is easy to traverse and switch to the library-based method only when traversal becomes difficult. Thus, in this paper we present a method that 1) Plans a guiding torso path which accounts for the difficulty of traversing the environment as predicted by learned regressors; and 2) Decomposes the guiding path into a set of segments, each of which is assigned a motion mode (i.e. a set of feet and hands to use) and a planning method. Easily-traversable segments are assigned a discrete-search planner, while other segments are assigned a library-based method that fits existing motion plans to the environment near the given segment. Our results suggest that this segmentation approach greatly outperforms standard discrete planning and that using the library-based method for more difficult segments gives a benefit over using discrete planning.

In this paper we develop a real-time human augmented mapping system. This approach replaces the traditional offline post processing of maps by a user-friendly system allowing for online editing capabilities. A wide number of applications that acquire accurate mapping of the environment could benefit from such a solution. The proposed framework consists of two main parts: 2D map building using LIDAR, encoders, and IMU; and a user interface for human map augmentation. The first part is built over Gmapping ROS package, while the second is developed in Unity software. Realworld experiments validated the ability of our system to correct for sensor noise and various mapping errors, thus increasing the accuracy of the obtained maps without additional computational costs.

Human-Robot shared workspace is a dynamic and unstructured environment. In such environment, pre-programmed robot paths might cause collisions or production disruptions. In order to solve this problem, motion planning framework has been proposed that adapts the robot's movement (path and speed)according to the human movement or any other dynamic obstacles in real time. Firstly, it defines the safety distance between robot and human or other obstacles. During run-time the 3D-Smart-Sensors capture the current position of human and other dynamic/static obstacles in human-robot shared workspace. The proposed framework plans and optimizes collision free robot trajectories with consideration for safety distance, path length and executing time. Once a new optimal trajectory is found, the framework controls the robot to adjust its movement paths. Moreover, the proposed framework can adjust the robot velocity based on the 3D-Zone Model of human-robot shared workspace. Therefore, the robot can reach its goal quickly and safely in a dynamic and unstructured environment.

This paper presents a system for cooperatively manipulating large objects between a human and a robot. This physical interaction system is designed to handle, transport, or manipulate large objects of different shapes in cooperation with a human. Unique points are the bi-manual physical cooperation, the sequential characteristic of the cooperation including contact changes, and a novel architecture combining force interaction cues, interactive search-based planning, and online trajectory and motion generation. The resulting system implements a mixed initiative collaboration strategy, deferring to the human when his intentions are unclear, and driving the task once understood. This results in an easy and intuitive human-robot interaction. It is evaluated in simulations and on a bi-manual mobile robot with 32 degrees of freedom.

Collaborative robotics over the last few years has gained increasing interest in the industrial scenario. Co-bots can be equipped with vision sensors and cognitive software layers, allowing the robot to figure out human intentions. To make this level of perception possible, human pose estimation algorithms are required. Several techniques have been already proposed to tackle this problem, which however present some weaknesses in particular when occlusions occur. This work proposes an algorithm for human pose estimation in the situations of partial occlusion, based on particle filter techniques. We have proved its validity in a realistic human-robot coexistence scenario, where a human and a dual arm robot have to perform tasks in a shared workspace.

Accurate long-term prediction of human motion in populated spaces is an important but difficult task for mobile robots and intelligent vehicles. What makes this task challenging is that human motion is influenced by a large variety of factors including the person's intention, the presence, attributes, actions, social relations and social norms of other surrounding agents, and the geometry and semantics of the environment. In this paper, we consider the problem of computing human motion predictions that account for such factors. We formulate the task as an MDP planning problem with stochastic policies and propose a weighted random walk algorithm in which each agent is locally influenced by social forces from other nearby agents. The novelty of this paper is that we incorporate social grouping information into the prediction process reflecting the soft formation constraints that groups typically impose to their members' motion. We show that our method makes more accurate predictions than three state-of-the-art methods in terms of probabilistic and geometrical performance metrics.

Intuitive exoskeleton control is fundamental since it contributes to improved user acceptance and wearability comfort. This requires the detection of user's motion intention and its incorporation into the exoskeleton control system. In this work, we propose a classification system based on Hidden Markov Models (HMMs), which facilitates the online classification of multi-modal sensor data acquired from a lower-limb exoskeleton based on previously defined motion patterns. For classification of these motion patterns at each time step, we consider the most recent sensor measurements by using a sliding window approach. We collected a training data set from a total number of 10 subjects performing 13 different motions with a passive exoskeleton equipped with 7 3D-force sensors and 3 inertial measurement units (IMUs). Our evaluation includes an analysis of the time needed for correct classification (latency), a validation for a training set containing all subjects and a leave-one-out validation to assess the generalization performance of the approach. The results indicate that our approach can classify motions of subjects included in the training set with an average accuracy of 92.80% and is able to achieve a generalization performance of 84.46%. With the selected parameters an average latency of 368.97 ms is achieved.

In human-robot collaboration, the robot is required to provide assistance to the user by facilitating task execution. However, due to stability requirements, a well-damped admittance behavior of the robot is necessary during interaction, thus inducing fatigue in the operator. While available schemes involve variable impedance controllers to mitigate this effect, here we propose an alternative approach entailing a proactive robot behavior that assists in the cooperative execution of trajectories towards desired goals, by estimating the user intention. To this end, we make use of Recurrent Neural Networks (RNNs) to predict and classify cooperative motions, on the basis of a set of predefined goals in the workspace and model-based generated data of human movements. Manual guidance validation experiments are conducted on a 6 d.o.f. ABB IRB140 industrial robot equipped with a force sensor.

Gaze provides subtle informative cues to aid fluent interactions among people. Incorporating human gaze predictions can signify how engaged a person is while interacting with a robot and allow the robot to predict a human's intentions or goals. We propose a novel approach to predict human gaze fixations relevant for human-robot interaction tasks-both referential and mutual gaze-in real time on a robot. We use a deep learning approach which tracks a human's gaze from a robot's perspective in real time. The approach builds on prior work which uses a deep network to predict the referential gaze of a person from a single 2D image. Our work uses an interpretable part of the network, a gaze heat map, and incorporates contextual task knowledge such as location of relevant objects, to predict referential gaze. We find that the gaze heat map statistics also capture differences between mutual and referential gaze conditions, which we use to predict whether a person is facing the robot's camera or not. We highlight the challenges of following a person's gaze on a robot in real time and show improved performance for referential gaze and mutual gaze prediction.

The words used to describe robotic performances include a degree of ambiguity that the human brain should solve without difficulty. However, the language used in -and about- robotics seems to escape from the ordinary processing of lexical ambiguity resolution. In this paper, we argue that there is no lack of an adequate language for robotics but that the lexicon at hand is forced by our representations. We investigate the main representational issues of the notions that express robotic actions and dispositions (i.e. behaviors).

Many scenarios require a robot to be able to explore its 3D environment online without human supervision. This is especially relevant for inspection tasks and search and rescue missions. To solve this high-dimensional path planning problem, sampling-based exploration algorithms have proven successful. However, these do not necessarily scale well to larger environments or spaces with narrow openings. This paper presents a 3D exploration planner based on the principles of Next-Best Views (NBVs). In this approach, a Micro-Aerial Vehicle (MAV)equipped with a limited field-of-view depth sensor randomly samples its configuration space to find promising future viewpoints. In order to obtain high sampling efficiency, our planner maintains and uses a history of visited places, and locally optimizes the robot's orientation with respect to unobserved space. We evaluate our method in several simulated scenarios, and compare it against a state-of-the-art exploration algorithm. The experiments show substantial improvements in exploration time (2 ⨯ faster), computation time, and path length, and advantages in handling difficult situations such as escaping dead-ends (up to 20 ⨯ faster). Finally, we validate the on-line capability of our algorithm on a computational constrained real world MAV.

This paper proposes an intelligent and fast tracking method for robust trackability against appearance changes. The method hybridizes a correlation-based tracking algorithm operating at hundreds of frames per second (fps) with a deep learning-based recognition algorithm operating at dozens of fps. A prototype intelligent mechanical tracking system was developed by implementing our hybridized tracking algorithm on a 500-fps vision platform. A complex-shaped target can be robustly tracked at the center of the camera view in real time by controlling a pan-tilt active vision system with 500 Hz visual feedback. The tracking performance of our proposed algorithm was verified by showing several experimental results for pre-learned objects, which were quickly manipulated against complex backgrounds.

Quickness, preciseness and robustness are required in manipulation tasks of robotic hands for automation of manufacturing sites. Previous researches have found that sensing from fingertips equipped with proximity sensors is available for the requirements, because it complements blind areas of vision sensors. In this paper, we develop a novel optical proximity sensor for robot fingertips which provides two levels of proximity information with different purposes, sampling rates, information quantity and quality. The lower-level information from the sensor is for high-speed feedback control of a robotic hand, and the higher-level information is for recognizing the shape and size of an object. A prototype of the sensor with 5 × 5 matrix of photo detectors is presented, and its availability is shown through basic characteristic tests. A motion experiment using a robotic hand equipped with the prototype sensors is also conducted. The result confirms that the robotic hand can adjust the position and orientation of the fingertips to various objects and then correct the grasping form according to the object size within 1s.

Stealth walking is a way of walking carefully and noiselessly, and is an approach to stable legged locomotion of underactuated robotic walkers on irregular terrains. This paper proposes a method for generating a high-speed stealth walking gait without including double-limb support phase, and discusses the effect of upper-body control and semicircular feet on the gait properties. First, we introduce a model of a 3-link planar underactuated biped with an upper body and semicircular feet, and derive the approximate target initial state of the upper body by using the linearized equation of motion. Second, we conduct numerical simulations of the nonlinear model to observe the typical stealth walking gaits, and analyze the changing tendency of the upper body motion with respect to the foot radius. Furthermore, we discuss the advantage of semicircular feet through parametric studies of the gait efficiencies.

When a disaster occurs, high output power should be available for rescue operation even if the electric supply is insufficient at site. This video presents a novel multi-fingered robot hand for extreme environments without a sufficient electric supply. The robot hand has four fingers with 16 joints and 12 degrees of freedom. The finger has a retention mechanism using no electrical power supply and a fingertip force of 150 [N]. Holding without power supply shows that our robot hand can lift a heavy barbell and keep its posture without using electrical power. The high fingertip force shows that steel cans can be crushed by the robot hand. In addition, dexterous motion of our robot hand shows that each finger allows flexion/extension and adduction/abduction. High-power manipulation shows that the robot hand can grasp and manipulate a hammer drill for making a hole in a concrete plate. The robot hand has a high potential for performing various tasks by obtaining high power output and electrical power saving.

This paper presents a hierarchical path planner consisting of two stages: a global planner that uses workspace information to create collision-free paths for the robot end-effector to follow, and multiple local planners running in parallel that verify the paths in the configuration space by expanding a task-space rapidly-exploring random tree (RRT). We demonstrate the practicality of our approach by comparing it with state-of-the-art planners in several challenging path planning problems. While using a single tree, our planner outperforms other single tree approaches in task-space or configuration space (C-space), while its performance and robustness are comparable to or better than that of parallelized bidirectional C-space planners.