IROS 2019 论文列表

Humans perceive continuous high-dimensional information by dividing it into significant segments such as words and units of motion. We believe that such unsupervised segmentation is also important for robots to learn topics such as language and motion. To this end, we previously proposed a hierarchical Dirichlet process-Gaussian process-hidden semi-Markov model (HDP-GP-HSMM). However, an important drawback to this model is that it cannot divide high-dimensional time-series data. Further, low-dimensional features must be extracted in advance. Segmentation largely depends on the design of features, and it is difficult to design effective features, especially in the case of high-dimensional data. To overcome this problem, this paper proposes a hierarchical Dirichlet process-variational autoencoder-Gaussian process-hidden semi-Markov model (HVGH). The parameters of the proposed HVGH are estimated through a mutual learning loop of the variational autoencoder and our previously proposed HDP-GP-HSMM. Hence, HVGH can extract features from high-dimensional time-series data, while simultaneously dividing it into segments in an unsupervised manner. In an experiment, we used various motion-capture data to show that our proposed model estimates the correct number of classes and more accurate segments than baseline methods. Moreover, we show that the proposed method can learn latent space suitable for segmentation.

Legged locomotion is a challenging task for learning algorithms, especially when the task requires a diverse set of primitive behaviors. To solve these problems, we introduce a hierarchical framework that can automatically learn to decompose complex locomotion tasks. A high-level policy issues commands in the form of a latent vector and also selects for how long the low-level policy will execute the latent command. Concurrently, the low-level policy uses the latent command and only the robot's on-board sensors to control the robot's actuators. Our approach allows the high-level policy to run at a lower frequency than the low-level one. We test our framework on a path-following task for a dynamic quadruped robot and we show that steering behaviors automatically emerge in the latent command space as low-level skills are needed for this task. We then show efficient adaptation of the trained policy to new tasks by transfer of the trained low-level policy. Finally, we validate the policies on a real quadruped robot. To the best of our knowledge, this is the first application of end-to-end hierarchical learning to a real robotic locomotion task.

A common strategy to deal with the expensive reinforcement learning (RL) of complex tasks is to decompose them into a collection of subtasks that are usually simpler to learn as well as reusable for new problems. However, when a robot learns the policies for these subtasks, common approaches treat every policy learning process separately. Therefore, all these individual (composable) policies need to be learned before tackling the learning process of the complex task through policies composition. Moreover, such composition of individual policies is usually performed sequentially, which is not suitable for tasks that require to perform the subtasks concurrently. In this paper, we propose to combine a set of composable Gaussian policies corresponding to these subtasks using a set of activation vectors, resulting in a complex Gaussian policy that is a function of the means and covariances matrices of the composable policies. Moreover, we propose an algorithm for learning both compound and composable policies within the same learning process by exploiting the off-policy data generated from the compound policy. The algorithm is built on a maximum entropy RL approach to favor exploration during the learning process. The results of the experiments show that the experience collected with the compound policy permits not only to solve the complex task but also to obtain useful composable policies that successfully perform in their corresponding subtasks.

Persistent surveillance of a target space using multiple unmanned aerial vehicles (UAVs) has multiple applications such as geographical surveys, air quality monitoring, and security monitoring. The limited onboard battery capacity challenges the continuous operation in these applications of persistent robots. We consider the problem for recharging persistent robots using mobile depots. The mobile depots collectively compute a set of tours to recharge all persistent robots with the minimum total cost. Compared to other works, the persistent UAVs are not required to detour to a static depot for energy replenishment such as recharging or battery swapping. We formulate this problem as a Generalized Multiple Depots Travelling Salesman Problem (GMDTSP) on a complete graph. A heuristic-based algorithm Multiple Depots Random Select (RSMD) is proposed to solve the recharging problem efficiently. The RSMD has proved to have an analytical constant upper bound in the worst-case scenario. We also propose a post-processing heuristic (RSMD-IM) to improve the solution quality further. We demonstrate the efficiency and effectiveness of our algorithm via benchmark on multiple instances from TSPLIB and GTSPLIB. The simulation results show that RSMD and RSMD-IM will perform significantly faster than the state of the art heuristic solver LKH with a loss of about 10% of solution quality.

Mobile manipulation planning commonly adopts a decoupled approach that performs planning separately on the base and the manipulator. While this approach is fast, it can generate sub-optimal paths. Another direction is a coupled approach jointly adjusting the base and manipulator in a high-dimensional configuration space. This coupled approach addresses sub-optimality and incompleteness of the decoupled approach, but has not been widely used due to its excessive computational overhead. Given this trade-off space, we present a simple, yet effective mobile manipulation sampling method, harmonious sampling, to perform the coupled approach mainly in difficult regions, where we need to simultaneously maneuver the base and the manipulator. Our method identifies such difficult regions through a low-dimensional base space by utilizing a reachability map given the target end-effector pose and narrow passage detected by generalized Voronoi diagram. For the rest of simple regions, we perform sampling mainly on the base configurations with a predefined joint configuration, accelerating the planning process. We compare our method with the decoupled and coupled approaches in six different problems with varying difficulty. Our method shows meaningful improvements experimentally in terms of time to find an initial solution (up to 5.6 times faster) and final solution cost (up to 17% lower) over the decoupled approach, especially in difficult scenes with narrow space. We also demonstrate these benefits with a real, mobile Hubo robot.

During suturing tasks performed with minimally invasive surgical robots, configuration singularities and joint limits often force surgeons to interrupt the task and re-grasp the needle using dual-arm movements. This yields an increased operator's cognitive load, time-to-completion and performance degradation. In this paper, we propose a haptic-guided shared control method for grasping the needle with the Patient Side Manipulator (PSM) of the da Vinci robot avoiding such issues. We suggest a cost function consisting of (i) the distance from robot joint limits and (ii) the task-oriented manipulability along the suturing trajectory. Evaluating the cost and its gradient on the needle grasping manifold allows us to obtain the optimal grasping pose for joint-limit and singularity free robot movements during suturing. We compute force cues and display them through the Master Tool Manipulator (MTM) to guide the surgeon towards the optimal grasp. As such, our system helps the operator to choose a grasping configuration that allows the robot to avoid joint limits and singularities during post-grasp suturing movements. We show the effectiveness of the proposed haptic-guided shared control method during suturing using both simulated and real experiments. The results illustrate that our approach significantly improves the performance in terms of needle re-grasping.

Robot-assisted cutting is considered an important task in several fields, such as robotic surgery, nuclear decommissioning, waste management, and manufacturing. Despite the complex dexterity requirements of cutting tasks, very simple mechanically-linked master-slave manipulators still dominate many of the above fields (e.g., nuclear robotics). Moreover, even when more dexterous manipulators are available (e.g., in robot-assisted surgery), the employed systems show little or no autonomy, delegating all control to the experience of the human operator. To ameliorate this situation, we present two haptic shared-control approaches for robotic cutting. They are designed to assist the human operator by enforcing different nonholonomic-like constraints representative of the cutting kinematics. To validate our approach, we carried out a human-subject experiment in a real cutting scenario. We compared our shared-control techniques with each other and with a standard haptic teleoperation scheme. Results show the usefulness of assisted control schemes in complex applications such as cutting. However, they also show a discrepancy between objective and subjective metrics.

Service robots will require several important manipulation skills, including the ability to accurately measure and pour liquids. Prior work on robotic liquid pouring has primarily focused on visual techniques for sensing liquids, but these techniques fall short when liquids are obscured by opaque or closed containers. This paper proposes a complementary method for liquid perception via haptic sensing. The robot moves a container through a series of tilting motions and observes the wrenches induced at the manipulator's wrist by the liquid's shifting center of mass. That data is then analyzed with a physics-based model to estimate the liquid's mass and volume. In experiments, this method achieves error margins of less than lg and 2mL for an unknown liquid in a 600mL cylindrical container. The model can also predict the viscosity of fluids, which can be used for classifying water, oil, and honey with an accuracy of 98%. The estimated volume is used to precisely pour 100mL of water with less than 4% average error.

Vascular diseases are the most common precursors to ischemic heart disease and stroke, which are two of the leading causes of death worldwide. Endovascular intervention is a minimally invasive surgical approach to treat such diseases. Compared to open surgery, it has the advantages of faster recovery, reduced need for general anesthesia, reduced blood loss and significantly lower mortality. Endovascular procedures require high surgical skills to minimize contacts between the manipulated instruments (catheters and guidewires) and the vessel wall, which represent one of the major risks for the patient. Robotic assistance can potentially improve the precision and stability of instruments manipulation. One key limitation of current commercial robotic platforms is the lack of haptic feedback, preventing their acceptance and limiting the clinical usability. This paper proposes to bring the benefit of haptic feedback to robot-assisted endovascular intervention. Here we hypothesize that the introduction of 3D haptic guidance during robot-assisted endovascular procedure can further improve the surgical performance and safety while overcoming the limitations of currently available technology. The proposed 3D haptic guidance allows the surgeon to sense the vasculature while controlling a catheter through a robotic haptic manipulator. Validation of the system is performed through end-user experiments with vascular surgeons on a bespoke surgical simulator. The obtained results demonstrate that 3D haptic guidance has the potential of improving effectiveness, precision, and safety of endovascular intervention. Furthermore, vascular surgeons found the proposed technology safe and overall easy to use, indicating its potential on real surgical procedures.

We develop a novel cube-shaped finger-wearable haptic interface named as HaptiCube. With the aim of implementing the tactile device with improved functionality and wearability, we focus on designing the device to have two characteristics: multi-DOF force feedback displayability and large force capability with compact and lightweight structure. In designing the device, we mainly consider the type and configuration of actuator and driving mechanism, since both have a great effect on the size, weight, and output force of the device. As the actuator, we select the shape memory alloy exhibiting high energy density. And, as the driving mechanism, two different types of compliant mechanism are designed for two important functions: to convert contraction of the SMA to the desired motion as a motion guide and to apply bias-force to the SMA as a bias-spring. The device is designed as the miniature interface that can display 3-DOF pressure and 2-DOF shearing force to user’s fingerpad. And, it is implemented as the functional prototype with the total weight of 26 g including actuators and all mechanical components. The experimental results on working performances (e.g. stroke, output force, and bandwidth) demonstrate that the device has superiority in terms of multi-DOF displayability, compact-sizability and wearability.

Recently, diverse research has actively been conducted to control the posture of jumping robots using an inertial tail mechanism. However, the inertial tail mechanism has a high probability of collision with obstacles. In this study, a momentum wheel mechanism is proposed to achieve the same attitude control performance while reducing the volume occupied by the inertial tail mechanism. To verify the performance of the momentum wheel mechanism, we proposed a jumping robot with a momentum wheel mechanism and performed a dynamic analysis, simulation, and experiments on a jumping robot with a momentum wheel mechanism. In addition, it has been demonstrated that the momentum wheel mechanism can contribute to control of the body angle of the jumping robot. As a result, the momentum wheel mechanism can enhance the stability of the jumping robot more than the tail mechanism, and the momentum wheel mechanism contributes to the attitude control of the body angle, which allows the jumping robot to perform continuous jumping.

We develop a system to disambiguate object instances within the same class based on simple physical descriptions. The system takes as input a natural language phrase and a depth image containing a segmented object and predicts how similar the observed object is to the object described by the phrase. Our system is designed to learn from only a small amount of human-labeled language data and generalize to viewpoints not represented in the language-annotated depth image training set. By decoupling 3D shape representation from language representation, this method is able to ground language to novel objects using a small amount of language-annotated depth-data and a larger corpus of unlabeled 3D object meshes, even when these objects are partially observed from unusual viewpoints. Our system is able to disambiguate between novel objects, observed via depth images, based on natural language descriptions. Our method also enables viewpoint transfer; trained on human-annotated data on a small set of depth images captured from frontal viewpoints, our system successfully predicted object attributes from rear views despite having no such depth images in its training set. Finally, we demonstrate our approach on a Baxter robot, enabling it to pick specific objects based on human-provided natural language descriptions.

In this paper, we present a novel grasp planning algorithm for unknown objects given a registered point cloud of the target from different views. The proposed methodology requires no prior knowledge of the object, nor offline learning. In our approach, the gripper kinematic model is used to generate a point cloud of each finger workspace, which is then filled with spheres. At run-time, first the object is segmented, its major axis is computed, in a plane perpendicular to which, the main grasping action is constrained. The object is then uniformly sampled and scanned for various gripper poses that assure at least one object point is located in the workspace of each finger. In addition, collision checks with the object or the table are performed using computationally inexpensive gripper shape approximation. Our methodology is both time efficient (consumes less than 1.5 seconds in average) and versatile. Successful experiments have been conducted on a simple jaw gripper (Franka Panda gripper) as well as a complex, high Degree of Freedom (DoF) hand (Allegro hand).

This paper presents a novel strategy for autonomous graph-based exploration path planning in subterranean environments. Attuned to the fact that subterranean settings, such as underground mines, are often large-scale networks of narrow tunnel-like and multi-branched topologies, the proposed planner is structured around a bifurcated local-and global-planner architecture. The local planner employs a rapidly-exploring random graph to reliably and efficiently identify collision-free paths that optimize an exploration gain within a local subspace. Accounting for the robot endurance limitations and the possibility that the local planner reaches a dead-end (e.g. a mine heading), the global planner is engaged when a return-to-home path must be derived or when the robot should be re-positioned towards an edge of the exploration space. The proposed planner is field evaluated in a collection of deployments inside both active and abandoned underground mines in the U.S. and in Switzerland.

There is an increasing interest in Reinforcement Learning to solve new and more challenging problems, as those emerging in robotics and unmanned autonomous vehicles. To face these complex systems, a hierarchical and multi-scale representation is crucial. This has brought the interest on Hierarchical Deep Reinforcement learning systems. Despite their successful application, Deep Reinforcement Learning systems suffer from a variety of drawbacks: they are data hungry, they lack of interpretability, and it is difficult to derive theoretical properties about their behavior. Classical Hierarchical Reinforcement Learning approaches, while not suffering from these drawbacks, are often suited for finite actions, and finite states, only. Furthermore, in most of the works, there is no systematic way to represent domain knowledge, which is often only embedded in the reward function. We present a novel Hierarchical Reinforcement Learning framework based on the hierarchical design approach typical of control theory. We developed our framework extending the block diagram representation of control systems to fit the needs of a Hierarchical Reinforcement Learning scenario, thus giving the possibility to integrate domain knowledge in an effective hierarchical architecture.

Mental simulation is a critical cognitive function for goal-directed behavior because it is essential for assessing actions and their consequences. When a self-generated or externally specified goal is given, a sequence of actions that is most likely to attain that goal is selected among other candidates via mental simulation. Therefore, better mental simulation leads to better goal-directed action planning. However, developing a mental simulation model is challenging because it requires knowledge of self and the environment. The current paper studies how adequate goal-directed action plans of robots can be mentally generated by dynamically organizing top-down visual attention and visual working memory. For this purpose, we propose a neural network model based on variational Bayes predictive coding, where goal-directed action planning is formulated by Bayesian inference of latent intentional space. Our experimental results showed that cognitively meaningful competencies, such as autonomous top-down attention to the robot end effector (its hand) as well as dynamic organization of occlusion-free visual working memory, emerged. Furthermore, our analysis of comparative experiments indicated that the introduction of visual working memory and the inference mechanism using variational Bayes predictive coding significantly improved the performance in planning adequate goal-directed actions.

Wall polishing is a typical and essential procedure in the interior renovation. However, such works are mainly carried out by humans, which have the disadvantages of low efficiency, inconsistent quality, and issues of safety and health. A new vision-based impedance controller is proposed for polishing robots to automate the labor-intensive works. The desired impedance model is specified as the control objective to regulate the dynamic relationship between the interaction force and the motion of the robot end effector, where the motion is measured with the vision feedback. The use of the vision feedback guarantees the performance of the robot from two aspect. First, the vision feedback from the high-resolution camera ensures the accuracy of measurement of the robot end effector and hence guarantees the quality of polishing. Second, the concept of image moment is introduced such that the image Jacobian matrix is non-singular in a global sense, which guarantees the large working range of the robot. The dynamic stability of the closed-loop system is rigorously proved with Lyapunov methods, and experimental results are presented to illustrate the performance of the proposed controller.

Transparent objects are prevalent across many environments of interest for dexterous robotic manipulation. Such transparent material leads to considerable uncertainty for robot perception and manipulation, and remains an open challenge for robotics. This problem is exacerbated when multiple transparent objects cluster into piles of clutter. In household environments, for example, it is common to encounter piles of glassware in kitchens, dining rooms, and reception areas, which are essentially invisible to modern robots. We present the GlassLoc algorithm for grasp pose detection of transparent objects in transparent clutter using plenoptic sensing. GlassLoc classifies graspable locations in space informed by a Depth Likelihood Volume (DLV) descriptor. We extend the DLV to infer the occupancy of transparent objects over a given space from multiple plenoptic viewpoints. We demonstrate and evaluate the GlassLoc algorithm on a Michigan Progress Fetch mounted with a first generation Lytro. The effectiveness of our algorithm is evaluated through experiments for grasp detection and execution with a variety of transparent glassware in minor clutter.

State-of-the-art methods for counting people in crowded scenes rely on deep networks to estimate crowd density in the image plane. While useful for this purpose, this image-plane density has no immediate physical meaning because it is subject to perspective distortion. This is a concern in sequences acquired by drones because the viewpoint changes often. This distortion is usually handled implicitly by either learning scale-invariant features or estimating density in patches of different sizes, neither of which accounts for the fact that scale changes must be consistent over the whole scene.In this paper, we explicitly model the scale changes and reason in terms of people per square-meter. We show that feeding the perspective model to the network allows us to enforce global scale consistency and that this model can be obtained on the fly from the drone sensors. In addition, it also enables us to enforce physically-inspired temporal consistency constraints that do not have to be learned. This yields an algorithm that outperforms state-of-the-art methods in inferring crowd density from a moving drone camera especially when perspective effects are strong.