IROS 2018 论文列表

For wheeled mobile systems, the wheel odometry is an important source of information about the current motion of the vehicle. It is used e.g. in the context of pose estimation and self-localization of planetary rovers, which is a crucial part of the success of planetary exploration missions. Depending on the wheel-soil interaction properties, wheel odometry measurements are subject to inherent errors such as wheel slippage. In this paper, a parameter-based approach for whole-body slip modeling and calibration is applied to a four-wheeled lightweight rover system. Details on the method for slip parameter calibration as well as the system-specific implementation are given. Experimental results from a test campaign on Mt. Etna are presented, showing significant improvements of the resulting wheel odometry measurements. The results are validated during a long range drive of approx. 900 m and discussed w. r. t. the advantages but also limitations of the method within a space exploration scenario.

In multi-finger or multi-arm grasping with friction contacts, maintaining force closure during motion is critical. Violation of this condition would cause contact slippage and possibly loss of grasp. This issue is of particular importance in space robotics, where the loss of grasp could lead to catastrophic consequences. There has been ample literature on stable grasp and force closure under static conditions. This paper investigates multi-arm grasping during motion, where the inertial force from the load could adversely affect grasp stability. Our approach dynamically adjusts the squeeze force and commanded robot/load motion to maintain a safe force closure condition. For a specified motion trajectory, the squeeze force is updated to prevent slippage based on the estimated inertial force. When the required squeeze force is beyond what the manipulators can safely apply, the trajectory will be scaled to reduce the inertial force component. In addition to motion-induced disturbance force, contact between the load and other objects in the environment can also cause slippage. The slip prevention strategy is extended to this case as well. The application scenario is based on the dual-arm transportation and berthing of a load in a micro-gravity environment. For laboratory testing, we use a fixed-base dual-arm robot to grasp, transport, and berth an object on a planar air bearing table. We also extend the transportation tests to a more general spatial setting, and use the dynamic squeeze adjustment to grasp, lift, and transport an object. Experimental results show the proposed method is effective at avoiding contact slippage during motion and when the object is in contact with the environment.

Continuum robots that move through undulatory actuation must be composed of body materials that can enable flexible movement yet also provide resistive forces to the surrounding fluid, granular, or solid environments. This need for “f1exible-yet-stiff” materials is notably important in robot designs that use passive propulsive elements such as tails and wings. Here we explore a laminate design paradigm for “f1exible-yet-stiff” robotic materials through sliding layer laminates (SLLs). We present design principles motivated by theory and experiment and illustrate a taxonomy of SLL enabled morphable materials capable of up to 7 fold change in stiffness. Lastly, we demonstrate the applicability of SLLs to undulatory continuum robots: a swimming robot with a passive tail. We target two desired robot locomotor behaviors: fast open water swimming, and steady swimming through narrow channels emulating underwater caverns and pipes. We demonstrate how tuning the stiffness of the robot tail maximizes thrust generation in these two locomotion modes. Soft tails are optimal in confined swimming because they generate short amplitude high wavenumber oscillations, while stiff tails in confined environments either collide with the walls or do not generate sufficient thrust. However, stiff tails are far better in unconfined environments which enable large stroke amplitudes requiring high stiffness. Through this demonstration we show that stiff or soft tail designs alone are incapable of effective locomotion in complex underwater environments challenge.

This communication presents an application for the use of ontologies in the generation of robot structures. The ontology developed for this app relies on the IEEE Standard Ontologies for Robotics and Automation (ORA) and it incorporates a set of concepts, relations and axioms that link robotic skills with the structural parts needed for their realization. The user can select a base configuration and/or a set of desired skills that the robot should be able to perform. Then, the application evaluates the axioms and returns an abstract structure that can carry out the requested skills. The final implementation of the structure can be achieved with any modular robotic platform that could identify each structural part with a physical device.

Traditional legged robots are capable of traversing challenging terrain, but lack of energy efficiency when compared to wheeled systems operating on flat environments. The combination of both locomotion domains overcomes the trade-off between mobility and efficiency. Therefore, this paper presents a novel motion planner and controller which together enable a legged robot equipped with skates to perform skating maneuvers. These are achieved by an appropriate combination of planned reaction forces and gliding motions. Our novel motion controller formulates a Virtual Model Controller and an optimal contact force distribution which takes into account the nonholonomic constraints introduced by the skates. This approach has been tested on the torque-controllable robot ANY mal equipped with passive wheels and ice skates as end-effectors. We conducted experiments on flat and inclined terrain, whereby we show that skating motions reduces the cost of transport by up to 80 % with respect to traditional walking gaits.

Research in human teamwork shows that a key element of fluid and fluent interactions is the interpretation of implicit verbal and non-verbal cues in context. This poses an issue to robotic platforms, however, as they have historically worked best when controlled through explicit commands that have employed structured, unequivocal representations of the external world and their human partners. In this work, we present a framework for effectively grounding situated and naturalistic speech to action selection during human-robot collaborative activities. This is accomplished by maintaining and incrementally updating separate “speech” and “context” models that jointly classify a collaborator's utterance. We evaluate the efficacy of the system on a collaborative construction task with an autonomous robot and human participants. We first demonstrate that our system is capable of acquiring and deploying new task representations from limited and naturalistic data sets, and without any prior domain knowledge of language or the task itself. Finally, we show that our system is capable of significantly improving performance on an unfamiliar task after a one-shot exposure.

Robots need to use their end-effectors not only to grasp and manipulate objects but also to understand the environment surrounding them. Object identification is of paramount importance in robotics applications, as it facilitates autonomous object handling, sorting, and quality inspection. In this paper, we present a new hyper-adaptive robot hand that is capable of discriminating between different everyday objects, as well as `model' objects with the same external geometry but varying material, density, or volume, with a single grasp. This work leverages all the benefits of simple, adaptive grasping mechanisms (robustness, simplicity, low weight, adaptability), a Random Forests classifier, tactile modules based on barometric sensors, and radar technology offered by the Google Soli sensor. Unlike prior work, the method does not rely on object exploration, object release or re-grasping and works for a wide variety of everyday objects. The feature space used consists of the Google Soli readings, the motor positions and the contact forces measured at different time instances of the grasping process. The whole approach is model-free and the hand is controlled in an open-loop fashion, achieving stable grasps with minimal complexity. The efficiency of the designs, sensors, and methods has been experimentally validated with experimental paradigms involving model and everyday objects.

This paper proposes a mixed-integer convex programming formulation for dynamic motion planning. Many dynamic constraints such as the actuator torque constraint are nonlinear and non-convex due to the trigonometrical terms from the Jacobian matrix. This often causes the optimization problem to converge to local optima or even infeasible set. In this paper, we convexify the torque constraint by formulating a mixed-integer quadratically-constrained program (MIQCP). More specifically, the workspace is discretized into a union of disjoint polytopes and torque constraint is enforced upon a convex outer approximation of the torque ellipsoid, obtained by solving a semidefinite program (SDP). Bilinear terms are approximated by McCormick envelope convex relaxation. The proposed MIQCP framework could be solved efficiently to global optimum and the generated trajectories could exploit the rich features of the rough terrain without any initial guess from the designer. The demonstrated experiment results prove that this approach is currently capable of planning consecutive jumps that navigates a single-legged robot through challenging terrains.

This work proposes the utilization of sinc functions as kernels of Dynamic Movement Primitives (DMP) models for encoding point-to-point kinematic behaviors. The proposed method presents a number of advantages with respect to the state of the art, as it (i) involves a simple learning technique, (ii) provides a method to determine the minimum required number of basis functions, based on the frequency content of the demonstrated motion and (iii) provides the ability to pre-define the reproduction accuracy of the learned behavior. The ability of the proposed model to accurately reproduce the behavior is demonstrated through simulations and experiments. Comparisons with the Gaussian-based DMP model show the proposed method's superiority in terms of computational complexity of learning and accuracy for a specific number of kernels.

We propose novel techniques for task allocation and planning in multi-robot systems operating in uncertain environments. Task allocation is performed simultaneously with planning, which provides more detailed information about individual robot behaviour, but also exploits independence between tasks to do so efficiently. We use Markov decision processes to model robot behaviour and linear temporal logic to specify tasks and safety constraints. Building upon techniques and tools from formal verification, we show how to generate a sequence of multi-robot policies, iteratively refining them to reallocate tasks if individual robots fail, and providing probabilistic guarantees on the performance (and safe operation) of the team of robots under the resulting policy. We implement our approach and evaluate it on a benchmark multi-robot example.

Arrangement of items on shelves in stores or warehouses is a tedious, repetitive task that can be feasible for robots to perform. The diversity of products that are available in stores and the different setups and preferences of each store makes pre-programming a robot for this task extremely challenging. Instead, our work argues for enabling end-users to customize the robot to their specific objects and setup at deployment time by programming it themselves. To that end, this paper contributes (i) a task representation for shelf arrangements based on a large dataset of grocery store shelf images, (ii) a method for inferring goal configurations from user inputs including demonstrations and direct parameter specifications, and (iii) a system implementation of the proposed approach that allows simultaneously learning task goals and actions. We evaluate our goal inference approach with ten different teaching strategies that combine alternative user inputs in different ways on the large dataset of grocery configurations, as well as with real human teachers through an online user study (N=32). We evaluate our full system implemented on a Fetch mobile manipulator on eight benchmark tasks that demonstrate end-to-end programming and execution of shelf arrangement tasks.

We report results of the analyses of the effect of communication through a humanoid robot in comparison with in-person, video-chat, and speaker on frontal brain activity of older people during an storytelling experiment. Our results suggest that whereas communicating through a physically embodied medium potentially induces a significantly higher pattern of brain activation with respect to video-chat and speaker, its difference is non-significant in comparison with in-person communication. These results imply that communicating through a humanoid robot induces a pattern of brain activity in older people that is potentially similar to in-person communication. Our findings benefit researchers and practitioners in rehabilitation and elderly care facilities in search of effective means of communication with their patients to increase their involvement in the incremental steps of their treatments. Moreover, they imply the utility of brain information as a promising sensory gateway in characterization of the behavioural responses in human-robot interaction.

Let's imagine a swarm of drones that has to visit some locations and build a map in a disaster area. Let's assume the drones only can communicate to their neighbors and manage partial information of the mission. A relevant question in this scenario is “Should the robots compete or should they cooperate?”. This work analyzes the described scenario to answer this question. Two game theoretical algorithms have been developed: one competitive and another cooperative. The competitive algorithm poses games among each drone and its neighbors and searches the Nash Equilibrium. The cooperative one defines electoral systems that allow the drones to vote their preferred task allocations for their neighbors. Both algorithms are extensively tested in multiple scenarios with different features. After the experiments the question can be answered “The robots should cooperate!”.

The goal of this research is to develop a lower-extremity wearable link mechanism (i.e., exoskeleton robot) that is capable of reducing load against targeted body parts such as bones, joints and muscles, for shock absorption that help to support exploration of extreme environments. One of the applications of such exoskeleton is to protect a pilot of a personal vertical mobility system, or JetPack, when landing. The shock absorbing exoskeleton is to introduce series and parallel viscoelasticity to the human skeletal system. The paper presents a pilot study to validate this body-protective exoskeleton concept by analyzing kinematic and dynamic models of a human-exoskeleton coupled system based on a multi-element viscoelastic model in rheology. A proof-of-concept prototype is developed and experimental data is presented.

Reinforcement learning is a promising approach to developing hard-to-engineer adaptive solutions for complex and diverse robotic tasks. However, learning with real-world robots is often unreliable and difficult, which resulted in their low adoption in reinforcement learning research. This difficulty is worsened by the lack of guidelines for setting up learning tasks with robots. In this work, we develop a learning task with a UR5 robotic arm to bring to light some key elements of a task setup and study their contributions to the challenges with robots11Source code of the task and the computational model behind the setup available at https://github.com/kindredresearch/SenseAct. We find that learning performance can be highly sensitive to the setup, and thus oversights and omissions in setup details can make effective learning, reproducibility, and fair comparison hard. Our study suggests some mitigating steps to help future experimenters avoid difficulties and pitfalls. We show that highly reliable and repeatable experiments can be performed in our setup, indicating the possibility of reinforcement learning research extensively based on real-world robots.

The Axel rappelling rover was designed to enable access to intriguing and important science sites that lie in difficult terrains that are inaccessible to conventional rovers. Extended autonomous rappelling calls for careful control of tether tension, precise management of tether spooling, and some measure of shock tolerance. This paper covers the design and testing of a first-generation tether management system (TMS) for Axel. The system uses a double bull-wheel capstan driven by a low-stiffness series elastic actuator (SEA) to provide tension control and decouple internal spooling tension from external tether tension. A series elastic actuator was chosen for this application to permit closed-loop tether tension control and to provide shock/drop tolerance of the rappelling system both while moving and when the system is inactive with the motors locked. Experiments on the new TMS show that this design performs well in keeping nearly constant spooling tension while rejecting large dynamic disturbances at the output. While the SEA is very effective at maintaining a given tension contribution, the additional effects of friction and the unique mechanical properties of the tether result in substantial errors in the measured output tension. Upcoming field trials will be used to evaluate the effectiveness and sufficiency of this system when integrated in Axel.

Humans recognize perceived continuous information by dividing it into significant segments such as words and unit motions. We believe that such unsupervised segmentation is also an important ability that robots need to learn topics such as language and motions. Hence, in this paper, we propose a method for dividing continuous time-series data into segments in an unsupervised manner. To this end, we proposed a method based on a hidden semi-Markov model with Gaussian process (GP-HSMM). If Gaussian processes, which are nonparametric models, are used, unit motion patterns can be extracted from complicated continuous motion. However, this approach requires the number of classes of segments in the time-series data in advance. To overcome this problem, in this paper, we extend GP-HSMM to a nonparametric Bayesian model by introducing a hierarchical Dirichlet process (HDP) and propose the hierarchical Dirichlet processes-Gaussian process-hidden semi-Markov model (HDP-GP-HSMM). In the nonparametric Bayesian model, an infinite number of classes is assumed and it becomes difficult to estimate the parameters naively. Instead, the parameters of the proposed HDP-GP-HSMM are estimated by applying slice sampling. In the experiments, we use various synthetic and motion-capture data to show that our proposed model can estimate a more correct number of classes and achieve more accurate segmentation than baseline methods.

Robots have replaced people in many manufacturing production lines but the information they gather from sensors might not be sufficient to autonomously accomplish dexterous manipulation operations. Symbiotic human-robot cooperation appears to be a more realistic near future in industrial scenarios. In this paper we present a configuration of human-robot collaboration in which the robot is sensory-augmented by means of a set of tactile signals coming from the human operator. The incorporation of low-level robot “intelligence” permits the cooperative manipulation of an object while enabling the human operator to stay focused on task itself and carry it out in the most natural way. The effectiveness of this approach is demonstrated in a use case in which a robot helps a human operator to successfully accomplish a writing task. System performance has been evaluated, considering several positions of the tiny vibration sensor in charge of gathering the human perception, by testing it on both the human hand and the co-manipulated object. Results suggest that the sensor provides valuable information for recognizing operator actions when it is placed either on the human hand or on the co-manipulated object. However, the sensor on the finger directly represents the operator's perception, while the output of the sensor attached to the object changes according to the distance between the interaction point and the sensor itself. In addition, in wearing the sensor, neither the object nor the robot need to be instrumented: the operator is free to interact with a large set of objects and collaborate with any existing robot without requiring supplemental equipment.

We demonstrate the physical rearrangement of wheeled stools in a moderately cluttered indoor environment by a quadrupedal robot that autonomously achieves a user's desired configuration. The robot's behaviors are planned and executed by a three layer hierarchical architecture consisting of: an offline symbolic task and motion planner; a reactive layer that tracks the reference output of the deliberative layer and avoids unanticipated obstacles sensed online; and a gait layer that realizes the abstract unicycle commands from the reactive module through appropriately coordinated joint level torque feedback loops. This work also extends prior formal results about the reactive layer to a broad class of nonconvex obstacles. Our design is verified both by formal proofs as well as empirical demonstration of various assembly tasks.