Attachments: Video Attachment
Keywords: Locomotion, Body balancing, Humanoid dynamics
Abstract: This paper proposes a fast computation method for 3D multi-contact locomotion. The contributions of this paper are (a) the derivation of the prospect centroidal dynamics by introducing a force distribution ratio, where the centroidal dynamics in multi-contact can be represented with a formulation similar to the inverted pendulum's one, (b) the development of a fast computation method for generating a 3D center of mass (CoM) trajectory. The centroidal dynamics becomes a linear time variant system (LTVS). The proposed method allows to generate a trajectory sequentially and to change the locomotion parameters at any time even under variable CoM height. Then the contact timing of the CoM motion can be adjusted to synchronize with the actual contact with environment by shortening or extending the desired duration of support phase. This can be used to improve the robustness of the locomotion. The validity of the proposed method is confirmed by several numerical results: a biped locomotion with variable height and a multi-contact motion.

Attachments: Video Attachment
Keywords: Locomotion, Body balancing, Locomotion planning
Abstract: We propose a methodology for dynamically balancing passive-ankled bipeds and full humanoids. As dynamic locomotion without ankle-actuation is more difficult than with actuated feet, our control scheme adopts an efficient whole-body controller that combines inverse kinematics, contact-consistent feed-forward torques, and low-level motor position controllers. To understand real-world sensing and controller requirements, we perform an uncertainty analysis on the linear-inverted-pendulum (LIP)-based footstep planner. This enables us to identify necessary hardware and control refinements to demonstrate that our controller can achieve long-term unsupported dynamic balancing on our series-elastic biped, Mercury. Through simulations, we also demonstrate that our control scheme for dynamic balancing with passive-ankles is applicable to full humanoid robots.

Attachments: Video Attachment
Keywords: Body balancing, Humanoid kinematics
Abstract: The spatial momentum relation of an underactuated articulated multibody system on a floating base is a dynamic equilibrium relation between its coupling and relative momenta. The relative momentum is the difference between the system momentum and the momentum of the composite-rigid-body (CRB) that is obtained when the joints are locked. This relation is referred to as the momentum equilibrium principle.

The focus in this work is on the angular momentum component of the momentum equilibrium principle. It is clarified that the relative angular momentum component can be represented in terms of the so-called relative angular velocity that is used as a control input in a balance controller. The balance controller proposed here is a whole-body controller that has independent inputs for center of mass (CoM) velocity and base-link angular velocity control. In addition, the relative angular velocity control input endows the controller with the unique property of generating an appropriate upper-limb motion that can stabilize the system momentum. More specifically, it is shown that when the relative angular velocity is derived from the reaction null-space (RNS) of the system, it becomes possible to stabilize the unstable states with a rolling foot/feet.

The formulation is simple and yet quite efficient - there is no need to modify the contact model to account for the transitions between the stable and unstable contact states. There is also no need to command the upper-limb motion directly. A few simulation examples are presented to demonstrate and discuss the properties of the controller.

Attachments: Video Attachment
Keywords: Humanoid dynamics, Trajectory planning, Locomotion
Abstract: We present a new method for multi-contact motion planning which efficiently encodes internal dynamics of the robot without needing to use full models. Our approach is based on a five-mass model which is formulated by Cartesian points instead of joint angles. We solve direct optimization problems which include distance constraints between these points, Newtonian equations and integration constraints. We consider a given rhythm of contact switches but leave the phase-timings and contact positions free inside the optimization to provide more flexibility. Due to simpler equations and sparser problem structures, we can achieve very short optimization times in the order of few hundred milliseconds, which make the method suitable for application of online model predictive control. Aside from contact position and time adjustment properties, we can include precise foothold regions and synthesize dynamic motions by taking internal dynamics and momentums into account.

Attachments: Video Attachment
Keywords: Trajectory planning, Body balancing, Locomotion planning
Abstract: This paper presents a multi-contact approach to generalized humanoid fall mitigation planning that unifies inertial shaping, protective stepping, and hand contact strategies. The planner optimizes both the contact sequence and the robot state trajectories. A high-level tree search is conducted to iteratively grow a contact transition tree. At each edge of the tree, trajectory optimization is used to calculate robot stabilization trajectories that produce the desired contact transition while minimizing kinetic energy. Also, at each node of the tree, the optimizer attempts to find a self-motion (inertial shaping movement) to eliminate kinetic energy. This paper also presents an efficient and effective method to generate initial seeds to facilitate trajectory optimization. Experiments demonstrate show that our proposed algorithm can generate complex stabilization strategies for a simulated robot under varying initial pushes and environment shapes.

Attachments: Video Attachment
Keywords: Locomotion
Abstract: This paper introduces a sensor system and motion that measure and compensate the error in position and orientation between robot’s end-effectors and the rungs of ladder while climbing a ladder. This error has been a threat to vertical ladder climbing for a four-limbed robot and may directly cause failure in climbing, thus must be compensated. In detail, our error compensation system consists of 8 proximity sensors (2 sensors for each end-effector) powered by lithium batteries and data is transmitted via wireless communication. With the system constructed, corresponding algorithms to measure and calculate the amount of error in position and orientation as well as motion planning of the robot to compensate the error are proposed. Additionally, countermeasures are also prepared to deal with undesired situations, such as communication in low quality and incorrect data returned from proximity sensors. Simulation results of comparison with and without error compensation are presented and experiment results of the real robot are given to validate the effectiveness of our proposed system. Finally, discussions about results are shown and expected prospective works are concluded.

Attachments: Video Attachment
Keywords: Locomotion planning, Tactile perception, Multimodal interaction
Abstract: Human bipedal balance during standing and walking depends on several receptors including the cutaneous receptors in the glabrous skin of the foot sole. It has been shown in human-involved studies that the different areas of the sole have distinct sensitivities and serve a different purpose in both walking and standing. In humanoid robotics, the feedback to keep balance is mainly achieved using force-torque sensors mounted at the robot's ankles. Although these sensors can accurately estimate the center of pressure of a foothold, they cannot provide information about the pressure shape of the footprint and therefore can miss ill terrain conditions during locomotion. In this paper, we present a biologically inspired sole skin sensor based on the robot skin developed at our lab. The robot skin can enhance and complement the ankle force-torque sensors used in balancing and walking controllers by providing additional information that a force-torque sensor cannot produce. This additional information can be used to reconstruct the supporting polygon and the pressure footprint online. We present a case study where a force-torque sensor fails to detect the terrain conditions while the skin succeeds and the information is used to re-plan the footstep position.

Attachments: Video Attachment
Keywords: Locomotion planning, Trajectory planning, Locomotion
Abstract: A common strategy to generate efficient locomotion movements is to split the problem into two consecutive steps: the first one generates the contact sequence together with the centroidal trajectory, while the second step computes the whole-body trajectory that follows the centroidal pattern. While the second step is generally handled by a simple program such as an inverse kinematics solver, we propose in this paper to compute the whole-body trajectory by using a local optimal control solver, namely Differential Dynamic Programming (DDP). Our method produces more efficient motions, with lower forces and smaller impacts, by exploiting the Angular Momentum (AM). With this aim, we propose an original DDP formulation exploiting the Karush-Kuhn-Tucker constraint of the rigid contact model. We experimentally show the importance of this approach by executing large steps walking on the real HRP-2 robot, and by solving the problem of attitude control under the absence of external contact forces.

Attachments: Video Attachment
Keywords: Locomotion planning, Locomotion
Abstract: This paper presents a novel walking gait generator that allows the successful traversal of moving support surfaces such as conveyor belts, moving plates and escalators. The gait generator previews all steps of a complete gait sequence, while providing efficient matrix-vector based computations. The moving support surfaces are explicitly taken into account for the trajectory design. Multiple successful simulations of walking on different non-stationary ground surfaces prove the high quality of the proposed walking gait generator.

Attachments: Video Attachment
Keywords: Locomotion, Trajectory planning, Humanoid dynamics
Abstract: This paper presents a feedback control methodology for 3D dynamic underactuated bipedal walking, that couples an actuated spring-loaded-inverted-pendulum (aSLIP) for forward walking and the passive Linear Inverted Pendulum (LIP) for lateral balancing. The applications of the reduced order models are twofold. First, we utilize aSLIP optimization to design optimal leg length and angle trajectories, and use the LIP dynamics to find desired boundary condition for lateral roll. Second, we present two feedback stabilization laws which are based on the reduced order models and applied on the full robot to stabilize the sagittal walking and lateral balancing separately. The ultimate feedback controller on the full order 3D walking robot is implemented via control Lyapunov function based Quadratic Programs (CLF-QPs). In particular, the reduced order models are used to approximate the underactuated dynamics and plan desired trajectories that are tracked via CLF-QPs. The end result is 3D underactuated walking, demonstrated in simulation on the bipedal robot Cassie.

Keywords: Locomotion, Humanoid dynamics, Body balancing
Abstract: We have realized a biped walking control based on the spatially quantized dynamics (SQD) which discretize a continuous system by a constant unit length along the walk direction. By using SQD, a prescribed sagittal kinematic pattern can be transformed into dynamically consistent robot motion in real-time. The lateral motion is generated by preview control which uses future ZMP online calculated by SQD in every control cycle. A successful biped walk of HRP-2Kai with fully stretched knees and long stride is reported by the proposed method.

Attachments: Video Attachment
Keywords: Locomotion planning, Body balancing, Humanoid kinematics
Abstract: To realize natural and fast bipedal walking, we mimic the ZMP trajectory from human walking data based on the concept of Divergent Component of Motion (DCM). This paper presents an omnidirectional gait generator that is lightweight and universal. In addition, we integrate DCM track- ing and foot placement adjustment into our control framework to ensure a stable walking motion and push recovery. The method proposed was validated in RoboCup2018 AdultSize soccer competition, where our robot WALKER+ walked on grass field and kept balance after the push from other robots.

Keywords: Humanoid kinematics, Medical, health and mental care, Prostheses & Ortheses
Abstract: Human-robot interaction (HRI) is moving towards the human-robot synchronization challenge. In robots like exoskeletons, this challenge translates to the reliable motion segmentation problem using wearable devices. Therefore, our paper explores the possibility of segmenting the motion reversals of a rigid-IMU cluster using screw-based invariants. Moreover, we evaluate the reliability of this framework with regard to the sensor placement, speed and type of motion. Overall, our results show that the screw-based invariants can reliably segment the motion reversals of a rigid-IMU cluster.

Keywords: Visual perception, Novel sensing mechanisms, Exoskeletons and assistive devices
Abstract: Semantic perception is a key enabler in robotics, which supposes a very resourceful and efficient manner of applying vision information for upper-level navigation and manipulation tasks. Given the challenges on specular semantics such as water hazards, transparent glasses and metallic surfaces, polarization imaging has been explored to complement the RGB-based pixel-wise semantic segmentation because it reflects surface characteristics and provides additional attributes. However, polarimetric measurements generally entail prohibitively expensive cameras and highly accurate calibrations. Inspired by the representation power of Convolutional Neural Networks (CNNs), we propose to predict polarization information from monocular RGB images, precisely per-pixel polarization difference. The core of our approach is a cluster of efficient deep architectures building on factorized convolutions, hierarchical dilations and pyramid representations, aimed to produce both semantic and polarimetric estimations in real time. Comprehensive experiments demonstrate the qualified accuracy on a wearable exoskeleton humanoid robot.

Attachments: Video Attachment
Keywords: Task planning, Multimodal interaction, Concept and strategy learning
Abstract: In many environments, robots have to handle partial observations, occlusions, and uncertainty. In this kind of setting, a partially observable Markov decision process (POMDP) is the method of choice for planning actions. However, especially in the presence of non-expert users, there are still open challenges preventing mass deployment of POMDPs in human environments. To this end, we present a novel approach that addresses both incorporating user objectives during task specification and asking humans for specific information during task execution; allowing for mutual information exchange. In POMDPs, the standard way of using a reward function to specify the task is challenging for experts and even more demanding for non-experts. We present a new POMDP algorithm that maximizes the probability of task success defined in the form of intuitive logic sentences. Moreover, we introduce the use of targeted queries in the POMDP model, through which the robot can request specific information. In contrast, most previous approaches rely on asking for full state information which can be cumbersome for users. Compared to previous approaches our approach is applicable to large state spaces. We evaluate the approach in a box stacking task both in simulations and experiments with a 7-DOF KUKA LWR arm. The experimental results confirm that asking targeted questions improves task performance significantly and that the robot successfully maximizes the probability of task success while fulfilling user-defined task objectives.

Keywords: Novel actuation mechanisms
Abstract: The high power ability of humanoid robot is desired for application of running or jumping motions, the new generation of humanoid robot formulate the need for a low-mass high-torque motor. A frameless motor has been designed by the Group, it is very important to calculate the thermal field of the motor and get the conclusion for the choice of the motor parameters. The magneto-thermal coupling analysis of the motor was carried out based on the thermal network by the iterative calculation. By using the equivalent thermal network method, the element’s copper loss and core loss is coupled into elements in thermal analysis by keeping the same mesh structure between magnetic and thermal analysis. Other losses such as air friction loss, rotor loss are included in the model. A temperature rise calculation program was written and different position temperature distribution was obtained. In the meantime, the steady-state temperature of BLDC was calculated by using the finite element method (FEM). The experimental results show that, the actual torque performance of the motor can reach the target of our design, at last, temperature calculation results obtained from two different methods were compared with experimental data, and the correctness of the calculation model is verified.

Keywords: Visual perception, Deep Learning
Abstract: Perception and cognition play important roles in intelligent robot research. Before interacting with the environment, such as tasks of grasping or manipulating, the robot need to understand and infer what to do and how to do it first. Based on this, our paper presents a new CNN architecture called Visual Manipulation Relationship Network (VMRN) to help robot detect targets and predict the manipulation relationships in real time, which ensures that the robot can complete tasks in a safe and reliable way. To implement end-to-end training and meet real-time requirements in robot tasks, we propose the Object Pairing Pooling Layer (OP2L) to help to predict all manipulation relationships in one forward process. Moreover, in order to train VMRN, we collect a dataset named Visual Manipulation Relationship Dataset (VMRD) consisting of 5185 images with more than 17000 object instances and the manipulation relationships between all possible pairs of objects in every image, which is labeled by the manipulation relationship tree. The experimental results show that the new network architecture can detect objects and predict manipulation relationships simultaneously and meet the real-time requirements in robot tasks.

Keywords: Task planning, Locomotion planning, Body balancing
Abstract: In this paper, we present an approach for generating a variety of whole-body motions for a humanoid robot. We extend the available Model Predictive Control (MPC) approaches for walking on flat terrain to plan for both vertical motion of the Center of Mass (CoM) and external contact forces consistent with a given task. The optimization problem is comprised of three stages, i. e. the CoM vertical motion, joint angles and contact forces planning. The choice of external contact (e. g. hand contact with the object or environment) among all available locations and the appropriate time to reach and maintain a contact are all computed automatically within the algorithm. The presented algorithm benefits from the simplicity of the Linear Inverted Pendulum Model (LIPM), while it overcomes the common limitations of this model and enables us to generate a variety of whole body motions through external contacts. Simulation and experimental implementation of several whole body actions in multi-contact scenarios on a humanoid robot show the capability of the proposed algorithm.

Attachments: Video Attachment
Keywords: Sensorimotor learning, Grasping and Manipulation, Concept and strategy learning
Abstract: Policy search reinforcement learning allows robots to acquire skills by themselves. However, the learning procedure is inherently unsafe as the robot has no a-priori way to predict the consequences of the exploratory actions it takes. Therefore, exploration can lead to collisions with the potential to harm the robot and/or the environment. In this work we address the safety aspect by constraining the exploration to happen in safe-to-explore state spaces. These are formed by decomposing target skills (e.g., grasping) into higher ranked sub-tasks (e.g., collision avoidance, joint limit avoidance) and lower ranked movement tasks (e.g., reaching). Sub-tasks are defined as concurrent controllers (policies) in different operational spaces together with associated Jacobians representing their joint-space mapping. Safety is ensured by only learning policies corresponding to lower ranked sub-tasks in the redundant null space of higher ranked ones. As a side benefit, learning in sub-manifolds of the state-space also facilitates sample efficiency. Reaching skills performed in simulation and grasping skills performed on a real robot validate the usefulness of the proposed approach.

Attachments: Video Attachment
Keywords: Task planning, Trajectory planning
Abstract: This paper presents a generalized method to evaluate risks associated with humanoid robots executing manipulation tasks. Risks are defined as the product of probability, the likelihood of an event occurring, and severity, the resulting magnitude of harm should it occur. Rather than try to reduce the probability of failure events to zero, the objective of this work is to allow an experienced operator/supervisor to define if some failures are worse than others. In doing so, this allows the operator to judge whether high risk motions are necessary for the task at hand. Utilizing NASA's humanoid robot Valkyrie, our framework is demonstrated in both simulation and on the physical robot, with a pick and place task. We show that our method is capable of predicting failures for given motions based on their calculated risk.

Attachments: Video Attachment
Keywords: Grasping and Manipulation, Novel mechanism design, Novel materials
Abstract: This paper presents a novel approach for modeling one-degree-of-freedom human metacarpophalangeal/interphalangeal joints based on a teeth-guided compliant cross-four-bar linkage. The proposed model allows developing self-adaptive anthropomorphic fingers able to be 3D printed in a single step without any accessories, except for simple tendon wiring after the printing process, using basic single-material additive manufacturing. Teethguided compliant cross-four-bar linkages as finger joints not only provide monolithic fabrication without assembly but also increase precision of anthropomorphic robot fingers by removing nonlinear characteristics, thus reducing the complexity of control for delicate grasping. Kinematic analysis of the proposed compliant finger joints is detailed and nonlinear finite element analysis results demonstrating their advantages are reported. A two-fingered underactuated hand with teeth-guided compliant cross-four-bar joints is also developed and qualitative discussion on grasping is conducted.

Attachments: Video Attachment
Keywords: Learning from demonstration, Skill modelling
Abstract: As humanoid robots become commonplace, learning and control algorithms must take into account the new challenges imposed by this morphology, in order to fully exploit their potential. One of the most prominent characteristics of such robots is their bimanual structure. Most research on learning bimanual skills has focused on the coordination between end-effectors, exploiting operational space formulations. However, motion patterns in bimanual scenarios are not exclusive to operational space, also occurring at joint level. Moreover, in addition to position, the end-effector orientation is also essential for bimanual operation. Here, we propose a framework for simultaneously learning constraints in configuration and operational spaces, while considering end-effector orientations, commonly overlooked in previous works. In particular, we extend the Task-Parameterized Gaussian Mixture Model (TP-GMM) with novel, Jacobian-based operators that address the foregoing problem. The proposed framework is evaluated in a bimanual task with the COMAN humanoid that requires the consideration of operational and configuration space motions.

Keywords: Visual perception, Locomotion planning, Locomotion
Abstract: Motivated by experiments showing that humans regulate their walking speed in order to improve localization performance, in this paper we explore the effects of walking gait on biped humanoid localization. We focus on step length as a proxy for speed and because of its ready applicability to current footstep planners, and we compare the performance of three different sparse visual odometry (VO) algorithms as a function of step length: a direct, a semi-direct and an indirect algorithm. The direct algorithm’s performance decreased the longer the step lengths, which along with the analysis of inertial and force/torque data, point to a decrease in performance due to an increase of mechanical vibrations. The indirect algorithm’s performance decreased in an opposite way, i.e., showing more errors with shorter step lengths, which we show to be due to the effects of drift over time. The semi-direct algorithm showed a performance in-between the previous two. These observations show that footstep planning could be used to improve the performance of VO algorithms in the future.

Attachments: Video Attachment
Keywords: Learning from demonstration, Sensorimotor learning, Physical interaction
Abstract: We propose a two-phase programming by demonstration (PbD) framework, which enables fast deployment of complex bi-manual assembly tasks. The first phase is a pre-learning phase, where the robot observes multiple task demonstrations performed by humans. Applying motion segmentation, it builds a rough plan of the task to be accomplished. Next phase is the policy refinement with incremental learning, performed by the kinesthetic guidance of the robot. In this phase, the robot already knows the rough task plan, so it can actively follow the pre-learned trajectories. The operator can arbitrarily modify the execution speed by simply pushing the robot along the demonstrated trajectory. Moreover, it can drive the robot forward and backward, and incrementally modify only those parts of the trajectory that need the refinement. During this phase, the robot estimates also the interaction forces and environmental compliance, which is needed for a robust and stable accomplished of assembly tasks in the exploitation phase. The benefit of this framework is in improved learning efficiency since the operator can concentrate only on the fine adjustment of the pre-learned trajectory. The robot optimizes its configuration from the data obtained in the pre-learning phase, which substantially facilitates the learning of kinematic redundant mechanisms and learning of bi-manual robot mechanisms. The proposed scheme was validated in a task where a bi-manual robot composed of two Kuka LWR-4 robot arms performs an assembly task.

Keywords: Novel mechanism design, Grasping and Manipulation, Grasp and motion planning
Abstract: The traditional underactuated fingers do not integrate the three grasping modes of coupling, parallel pinching and self-adaption grasping. In order to solve this problem, this paper proposes coupling-parallel-adaption merged (CPAM) grasping mode. By realizing the switching of two states of coupling and parallel pinching, the coupling, parallel pinching and self-adaptive grasping modes are integrated into one finger. According to CAPM mode, this paper develops CPAM finger, which includes a power input device, a power transmission mechanism, a coupling transmission mechanism, a parallel-pinching transmission mechanism, a limiting mechanism and a switching device. The switching device is a gear plate mechanism, which can freely switch the coupling mode and parallel-pinching mode, and can keep the finger posture when switching. Theoretical analysis and experimental results show that CPAM fingers can grasp in a suitable mode according to the position, shape and size of the objects, and the grasping range is large

Keywords: Physical interaction, Skill modelling, Humanoid dynamics
Abstract: In future, robots will be present in everyday life. The development of these supporting robots is a challenge. A fundamental task for assistance robots is to pick up and hand over objects to humans. By interacting with users, soft factors such as predictability, safety and reliability become important factors for development. Previous works show that collaboration with robots is more acceptable when robots behave and move human-like. In this paper, we present a motion model based on the motion profiles of individual joints. These motion profiles are based on observations and measurements of joint movements in human-human handover. We implemented this joint motion model (JMM) on a humanoid and a non-humanoidal industrial robot to show the movements to subjects. Particular attention was paid to the recognizability and human similarity of the movements. The results show that people are able to recognize human-like movements and perceive the movements of the JMM as more human-like compared to a traditional model. Furthermore, it turns out that the differences between a linear joint space trajectory and JMM are more noticeable in an industrial robot than in a humanoid robot.

Attachments: Video Attachment
Keywords: Body balancing, Humanoid dynamics
Abstract: A balance stabilizer is proposed that has the capability of absorbing high-energy collisions without reactive stepping. The stabilizer is based on the spatial dynamics formulation and has the unique feature that the trunk rotation can be specified in an independent way from the desired rate of change of the system (centroidal) angular momentum. The formulation is based on the momentum equilibrium principle for floating-base robots and the relativity of angular momentum revealed in the companion paper. The stabilizer injects angular momentum damping via the so-called relative angular acceleration (RAA) derived from the reaction null-space (RNS) of the system. The damping is used to increase the robustness of the balance stabilizer at critical states such as foot roll. It is shown how to embed the RAA stabilizer into a joint-torque controller whereby the motion and force optimization tasks are solved in a single step, yielding a formulation that does not rely upon a general solver. The performance of the controller is examined via simulations whereby external impact-type disturbances are applied to the robot. One part of the impact energy is accommodated via the trunk rotations by lowering the respective PD feedback gains immediately after impact onset. It is then dissipated with higher gains, while recovering the stability of the posture. Another part of the impact energy yields foot roll; this part is dissipated with the angular momentum damping realized through an appropriate arm motion. When in a single stance, the angular momentum damping control yields a movement in the swing leg in addition to that in the arms. The motion in the leg injects additional angular momentum damping, such that a high-energy impact can be accommodated that would otherwise require a reactive stepping.

Keywords: Humanoid dynamics, Modelling and simulating humans, Locomotion
Abstract: Gait design for a biped robot is an intriguing problem. The objective is to replicate an efficient gait according to the jogging dynamics of a human in a biped robot. This paper aims to find an optimal gait for jogging dynamics of a biped robot on a continuous-time nonlinear mathematical model. The nonlinear model is approximated using the describing function method and requires the gait to be sinusoidal. It is revealed that the natural oscillation of an undamped biped robot is also an optimal gait. The optimal frequency reduces to compensate for damping. The characteristic of the optimal gait is further studied in extensive simulations.

Keywords: Modelling and simulating humans, Skill modelling, Grasp and motion planning
Abstract: Predicting human motion in unstructured and dynamic environments is challenging. Human behavior arises from complex sensory-motor couplings processes that can change drastically depending on environments or tasks. In order to alleviate this issue, we propose to encode the lower level aspects of human motion separately from the higher level geometrical aspects using data driven dynamical models. In order to perform longer term behavior predictions that account for variation in tasks and environments, we propose to make use of gradient based constraint motion optimization. The present method is the first to our knowledge to combine motion optimization and data driven dynamical models for human motion prediction. We present results on synthetic and motion capture data of upper body reaching movements that demonstrate the efficacy of the approach with respect to simple baselines often mentioned in prior work.

Attachments: Video Attachment
Keywords: Locomotion planning
Abstract: In this paper we propose an anytime planning/replanning algorithm aimed at generating motions allowing a humanoid to fulfill an assigned task that implicitly requires stepping. The algorithm interleaves planning and execution intervals: a previously planned whole-body motion is executed while simultaneously planning a new solution for the subsequent execution interval. At each planning interval, a specifically designed randomized local planner builds a tree in configuration-time space by concatenating successions of CoM movement primitives. Such a planner works in two stages. A first lazy stage quickly expands the tree, testing only vertexes for collisions; then, a second validation stage searches the tree for feasible, collision-free whole-body motions realizing a solution to be executed during the next planning interval. We discuss how the proposed planner can avoid deadlock and we propose how it can be extended to a sensor-based planner. The proposed method has been implemented in V-REP for the NAO humanoid and successfully tested in various scenarios of increasing complexity.

Attachments: Video Attachment
Keywords: Visual perception, Multimodal perception, Social interaction and acceptability
Abstract: The general ability to analyze and classify the 3D kinematics of the human form is an essential step in the development of socially adept humanoid robots. A variety of different types of signals can be used by machines to represent and characterize actions such as RGB videos, infrared maps, and optical flow. In particular, skeleton sequences provide a natural 3D kinematic description of human motions and can be acquired in real time using RGB+D cameras. Moreover, skeleton sequences are generalizable to characterize the motions of both humans and humanoid robots. The Globally Optimal Reparameterization Algorithm (GORA) is a novel, recently proposed algorithm for signal alignment in which signals are reparameterized to a globally optimal universal standard timescale (UST). Here, we introduce a variant of GORA for humanoid action recognition with skeleton sequences, which we call GORA-S. We briefly review the algorithm's mathematical foundations and contextualize them in the problem of action recognition with skeleton sequences. Subsequently, we introduce GORA-S and discuss parameters and numerical techniques for its effective implementation. We then compare its performance with that of the DTW and FastDTW algorithms, in terms of computational efficiency and accuracy in matching skeletons. Our results show that GORA-S attains a complexity that is significantly less than that of any tested DTW method. In addition, it displays a favorable balance between speed and accuracy that remains nearly invariant under changes in skeleton sampling frequency, lending it a degree of versatility that could make it well-suited for a variety of action recognition tasks.

Keywords: Body balancing, Humanoid dynamics
Abstract: In this paper we extend the line of research that aims at applying linear optimal control approaches with quadratic cost (LQR) to the inherently non-linear control problem of whole-body balancing for push recovery of humanoid robots. The non-linearity of the system is addressed in the controller design by optimization in the weight-space of the cost function in order to maximize balancing performance. We use stochastic sampling-based, gradient-free optimization over the large design parameter space of the whole-body controller to efficiently cope with the unknown relation between the cost function and the balancing performance. We further investigate three different linear ground contact models and evaluate their influence on the overall controller performance. We demonstrate that parameter optimization and novel ground contact models can be used to design a linear balancing controller that produces human-like whole-body motions in physics simulation-based push recovery experiments, simultaneously considering joint angles, center of mass and angular momentum.

Keywords: Body balancing, Humanoid dynamics, Novel mechanism design
Abstract: This paper presents a computational efficient balance control algorithm developed for a lightweight biped. A LIP model of the robot is combined with the ZMP calculation to derive a joint space control action based on a PD controller. Furthermore, a method is implemented to estimate the ZMP directly from the center of pressure measured using the force sensors installed under the feet of the robot. This, allows a real time implementation of the controller without using the robot direct kinematics, reducing model inaccuracies and improving the controller reactivity. Simulation results and tests on the real robot prototype shows that the control system is able to compensate for external disturbances forces up to 10N reducing the oscillations of 60%.

Keywords: Locomotion, Locomotion planning, Humanoid dynamics
Abstract: Energy consumption for bipedal walking plays a central role for a humanoid robot with limited battery capacity. Studies have revealed that exploiting the allowable Zero Moment Point region (AZR) and Center of Mass (CoM) height variation (CoMHV) are strategies capable of improving energy performance. In general, energetic cost is evaluated by integrating the electric power of multi joints. However, this Joint-Power-based Index requires computing joint torques and velocities in advance, which usually requires time-consuming iterative procedures, especially for multi-joints robots. In this work, we propose a CoM-Acceleration-based Optimal Index(CAOI) to synthesize an energetically efficient CoM trajectory. The proposed method is based on the Linear Inverted Pendulum Model, whose energetic cost can be easily measured by the input energy required for driving the point mass to track a reference trajectory. We characterize the CoM motion for a single walking cycle and define its energetic cost as Unit Energy Consumption. Based on the CAOI, an analytic solution for CoM trajectory generation is provided. Hardware experiments demonstrated the computational efficiency of the proposed approach and the energetic benefits of exploiting AZR and CoMHV strategies.

Attachments: Video Attachment
Keywords: Software and hardware architecture, Modelling and simulating humans, Humanoid dynamics
Abstract: Motivated by the similar structure and actuation mechanism to humans and animals, the study of musculoskeletal robots has gained attention in recent years. However, the unilateral actuation property of muscles and high complexity of the mechanical system imposes great challenges on the design and control of such robots. An open-source realistic simulation platform for theoretical testing would therefore be advantageous for the research community of musculoskeletal robots. In this paper, an end-to-end open-source framework for the design, simulation and control of the general class of musculoskeletal robots (CARDSFlow) is presented. The framework consists of three advantageous features: 1) 3D computer-aided designs of musculoskeletal robots can be automatically converted for use with Gazebo and CASPR; 2) realistic physics simulation of musculoskeletal robots within Gazebo; and 3) integration of CASPR cable-driven robot controllers with CARDSFlow through the ROS platform. Simulation results on a two-link planar robot and the Roboy robot arm are presented to demonstrate the convenience to design and simulate musculoskeletal robots using CARDSFlow.

Attachments: Video Attachment
Keywords: Locomotion planning, System integration, Locomotion
Abstract: In this paper, we propose a locomotion planning framework for a humanoid robot with an efficient footstep and whole-body collision avoidance planning, which enables the robot to traverse an unknown narrow space while utilizing its body structure like a human. The key idea of the proposed method is to reduce a large computational cost for the whole-body locomotion planning by executing global footstep planning first, which has a much smaller search space, and then performing a sequential whole-body posture planning while utilizing the resulting footsteps and a centroidal trajectory as a guide. In the global footstep planning phase, we modify bounding box of the robot based on the centroidal sway motion. This idea enables the planner to obtain appropriate footsteps for next whole-body motion planning. Then, we execute sequential whole-body collision avoidance motion planning by prioritized inverse kinematics based on the resulting footsteps and centroidal trajectory, which enables the robot to plan whole-body collision avoidance motion for each step within less than 100ms at worst. The major contribution of our paper is solving the problem of the increasing computational cost for whole-body motion planning and enabling a humanoid robot to execute adaptive locomotion planning on the spot in an unknown narrow space.

Attachments: Video Attachment
Keywords: System integration, Humanoid dynamics
Abstract: In this paper, we present a real-time method for identification of the inertial parameters of a humanoid robot and an estimation of its center-of-mass (CoM) and linear and angular momentum. CoM and momentum are important for whole-body motion generation of a humanoid robot and can be used as an indicator of motion planning. Because they are affected by modeling errors and inertia changes (e.g., due to object grasping), it is important to estimate them online. The proposed method has the advantage of being based only on the internal sensors. We verified the effectiveness of the proposal method by applying it to a humanoid robot.

Attachments: Video Attachment
Keywords: Grand challenges, competitions, Home, field, space, underwater, System integration
Abstract: Most field robots today work under partial or complete guidance of an operator. The operator monitors, or at times augments, the control inputs of the robot to achieve better results or desired behavior. Robots that are operated remotely and over low bandwidth channels limit the involvement of the operator, leaving them vulnerable to unanticipated scenarios. The NASA Space Robotics Challenge (SRC), held in 2016-17, posed a challenge to operate a simulated Valkyrie R5 humanoid robot over a minimum bandwidth of 64-4k bits/second uplink, 50k-380k bits/second downlink, and a maximum latency of 20 seconds. To achieve this, we designed and implemented extended state machines that allow a robot to perform known tasks autonomously in a partially known environment along with the flexibility to perform system critical interventions manually, if required. The main intuition behind our approach is to combine (a) sensor data redundancy for object detection and (b) 2-stage motion planning approach using state machines to successfully accomplish complex tasks. The complex tasks demonstrated are aligning a communication dish, picking up a solar panel, and deploying solar panels autonomously. The overall system design allowed successful completion of tasks even after sub-task failures and/or complete communication loss.

Attachments: Video Attachment
Keywords: Grasp and motion planning, Task planning
Abstract: This paper presents a Center of Mass (CoM) based manipulation and regrasp planner that implements stability constraints to preserve the robot balance. The planner provides a graph of IK-feasible, collision-free and stable motion sequences, constructed using an energy based motion planning algorithm. It assures that the assembly motions are stable and prevent the robot from falling while performing dexterous tasks in different situations. Furthermore, the constraints are also used to perform an RRT-inspired task-related stability estimation in several simulations. The estimation can be used to select between single-arm and dual-arm regrasping configurations to achieve more stability and robustness for a given manipulation task. To validate the planner and the task-related stability estimations, several tests are performed in simulations and real-world experiments involving the HRP5P humanoid robot, the 5th generation of the HRP robot family. The experiment results suggest that the planner and the task-related stability estimation provide robust behavior for the humanoid robot while performing regrasp tasks.

Attachments: Video Attachment
Keywords: Learning from demonstration, Skill modelling, Physical interaction
Abstract: Numerous robotics tasks involve complex physical interactions with the environment, where the role of variable impedance skills and the information of contact forces are crucial for successful performance. The dynamicity of our environments demands robots to adapt their manipulation skills to a large variety of situations, where learning capabilities are necessary. In this context, we propose a framework to teach a robot to perform manipulation tasks by integrating force sensing and variable impedance control. This framework endows robots with force-based variable stiffness skills that become relevant when vision information is unavailable or uninformative. Such skills are built on stiffness estimations that are computed from human demonstrations, which are then used along with sensed forces, to encode a probabilistic model of the robot skill. The resulting model is later used to retrieve time-varying stiffness profiles. We study two different stiffness representations based on emph{(i)} Cholesky decomposition, and emph{(ii)} Riemannian manifolds. For validation, we use a simulation of a 2D mass-spring-damper system subject to external forces, and a real experiment where a 7-DoF robot learns to perform a valve-turning task by varying its Cartesian stiffness.

Keywords: Multimodal perception, System integration, Prostheses & Ortheses
Abstract: The development of dexterous and robust anthropomorphic hands with rich sensor feedback remains a challenging task for both humanoid robotics as well as prosthetics as of today. The design of hands that are scalable in size and equipped with integrated multimodal sensor systems is a key requirement for advanced control schemes and reactive behaviour. In this paper, we present the design of a scalable and low cost robotic finger with a soft fingertip and position, temperature as well as normal and shear force sensors. All cables and sensors are completely enclosed inside the finger to ensure an anthropometric appearance. The finger is modelled based on a 50th percentile male little finger and can be easily adapted to other dimensions in terms of size and sensor system configuration. We describe the design of the sensor system, provide an experimental analysis for the characterization of the different sensor types in terms of sensor range, resolution, creep, spatial response as well as temperature flux.

Keywords: Humanoid dynamics, Home, field, space, underwater, Locomotion
Abstract: This paper proposes a control method for biped robot locomotion on uneven and uncertain terrain based on the impedance control with impedance modulation at the ankle joints and terrain recognition using force sensors at the soles. The impedance parameters are changed depending on unexpected contact forces. To reduce the size of the peak ground reaction force and to guarantee soft footing on the ground, the stiffness coefficient of the impedance of the landing foot is drastically reduced. Also, the orientation of landing foot is updated for the next phase of walking trajectory. A series of computer simulations of a 12-degree-of-freedom (DOF) biped robot with an uneven and uncertain terrain showed the effectiveness of the proposed control method.

Attachments: Video Attachment
Keywords: Grasping and Manipulation
Abstract: Consider the task of grasping the handle of a door, and then pushing it until the door opens. These two fundamental robotics problems (selecting secure grasps of a hand on an object, e.g. the door handle, and planning collision-free trajectories of a robot arm that will move that object along a desired path) have predominantly been studied separately from one another. Thus, much of the grasping literature overlooks the fundamental purpose of grasping objects, which is typically to make them move in desirable ways. Given a desired post-grasp trajectory of the object, different choices of grasp will often determine whether or not collision-free post-grasp motions of the arm can be found, which will deliver that trajectory. We address this problem by examining a number of possible stable grasping configurations on an object. For each stable grasp, we explore the motion space of the manipulator which would be needed for post-grasp motions, to deliver the object along the desired trajectory. A criterion, based on potential fields in the post-grasp motion space, is used to assign a collision-cost to each grasp. A grasping configuration is then selected which enables the desired post-grasp object motion while minimising the proximity of all robot parts to obstacles during motion. We demonstrate our method with peg-in-hole and pick-and-place experiments in cluttered scenes, using a Franka Panda robot. Our approach is effective in selecting appropriate grasps, which enable both stable grasp and also desired post-grasp movements without collisions. We also show that, when grasps are selected based on grasp stability alone, without consideration for desired post-grasp manipulations, the corresponding post-grasp movements of the manipulator may result in collisions.

Keywords: Tactile perception, Deep Learning, Home, field, space, underwater
Abstract: In this paper, we show that a robot equipped with a flexible and commercially available tactile skin can exceed human performance in the challenging tasks of material classification, i.e., uniquely identifying a given material by touch alone, and of material differentiation, i.e., deciding if the materials in a given pair of materials are the same or different. For processing the high dimensional spatio-temporal tactile signal, we use a new tactile deep learning network architecture {tactnetII} which is based on {tactnet}~cite{Baishya2016} and is significantly extended with recently described architectural enhancements and training methods. TactNet-II reaches an accuracy for the material classification task as high as 95.0.%. For the material differentiation a new Siamese network based architecture is presented which reaches an accuracy as high as 95.4.%. All the results have been achieved on a new challenging dataset of 36 everyday household materials.

In a thorough human performance experiment with 15 subjects, we show that the human performance is significantly lower than the robot's performance for both tactile tasks.

Attachments: Video Attachment
Keywords: Prostheses & Ortheses, Novel actuation mechanisms, Teleoperation
Abstract: Traditionally, robotic hand grasping is realized by rigid robotic hands or grippers, which requires high-resolution sensor feedback and delicate control algorithm. Recently, soft robotics has emerged as an alternative approach to humanoid robotic hand design. But due to distinctive material, structure, actuation mechanism, limited degrees-of-freedom (DOF) of soft robots, their control raised new challenges. Most existing soft robot control strategies are based on the simple on/off signal, rather than intuitive, real-time control for dexterous grasping and manipulation tasks. In this paper, we present an intuitive grasping control for our proprietary 13-DOF humanoid soft robotic hand, BCL-13. This control approach allows all the 13 independent DOFs to be controlled continuously by intuitive human hand poses. Real-time human hand joint angles are captured by Leap Motion Controller. Then the human hand joint angle position is mapped into the robotic hand joint through our dedicated filter. Finally, the robotic hand joint actuation commands are regulated by the lower-level pressure controller. With passive compliance, the proposed intuitive grasping process can achieve excellent grasping performance and safety without strict accuracy requirements. This approach shows potential for dexterous humanoid robotic hand control for safe and intuitive interactions.

Attachments: Video Attachment
Keywords: Body balancing, Humanoid dynamics, Physical interaction
Abstract: In this work we present a novel method to address the balancing problem for torque controlled legged robots through post-optimization of contact forces. The main concept consists in treating a legged robot as a fully actuated fixed-base system in order to compute the desired joint torques according to a previous work by the authors. The under-actuated component of the obtained torques is then mapped into contact forces through an optimal distribution problem. Besides extending our previous work to the floating-base case, the proposed method has the notable advantage of avoiding the specification of a desired momentum of rotation, in addition to a reduced number of decision variables compared to full-inverse dynamics methods. The effectiveness of our approach has been validated in simulation using two different humanoid platforms: the CENTAURO and the COMAN+ robots, both recently developed at Istituto Italiano di Tecnologia (IIT). Preliminary experimental results on COMAN+ are also presented.

Attachments: Video Attachment
Keywords: Grasping and Manipulation, Modelling and simulating humans
Abstract: Building anthropomorphic robotic and prosthetic hands is a challenging task due to size and performance requirements. As of today it is impossible for such artificial hands to mimic the capabilities of a human hand. A popular approach to reduce the complexity in hand design is the realization of hand synergies through underactuated mechanism, leading also to a reduction of control complexity. In this paper we aim to find grasp synergies of human grasps by employing a deep autoencoder. We perform a grasp study with 15 subjects including 2250 grasps on 35 diverse objects. The emerging latent space contains a comprehensive representation of grasp type and the size of the grasped object, while preserving a large amount of grasp information. In addition we report on novel findings on couplings and grasp specific features of joint kinematics, which can be directly applied to the control of anthropomorphic hands.

Keywords: Humanoid dynamics, Novel mechanism design, Medical, health and mental care
Abstract: To assist the elderly people and the patients with neurologic injuries for rehabilitation, the robot-assisted therapy is one of the most remarkable methods for this purpose. In this paper, we developed an exoskeleton based on Pneumatic Muscle Actuators (PMAs). By describing characteristics of human walking, a novel design was proposed to improve the walking comfort of the wearer. In addition, the dynamics of the exoskeleton were analyzed and divided into three parts: the modeling of PMAs, antagonistic configuration of PMAs and the mechanical structure of exoskeletons. Based on this model, a simulation platform was established. Furthermore, a model-free control strategy was utilized to get the exoskeleton properly controlled, which is called Proxy-based Sliding Mode Control. The effectiveness of proposed dynamical model was verified through comparison study of simulations and corresponding experiments.

Keywords: Modelling and simulating humans, Humanoids for human science, Humanoid dynamics
Abstract: This paper gives some macroscopic understandings on human walking about the limitations on walking speed and step length, the reachable region, capture region, and disturbance recovery through a 2D inverted pendulum model. Our concern is the most basic problems in human walking, such as what are the limitations on walking speed and step length, how people change speed during step-to-step transition, and how people prevent a fall. The concept of walking orbit is proposed as a tool to study these problems. It describes the walking motion in the state space under walking constraints, giving us an intuitive way to study human walking during a step and switch between steps. The model has a point mass on the hip and two massless legs. The two dominant control inputs, hip and ankle actuation are idealized into a free determined foot placement and an impulsive push off. Based on this model, some quantitative and qualitative analysis are given, leading to some macroscopic understandings on human walking. Although this paper does not talk about any details on how to realize the control for a real biped robot, it may serve as a helpful guide for biped robot design and control in the future.

Keywords: Exoskeletons and assistive devices, Multimodal perception
Abstract: To adapt to different locomotion modes or terrains, real-time human intents recognition is an essential skill to the control of lower-limb exoskeletons timely and precisely. In this paper, we propose a real-time on-board training and recognition method to identify locomotion-related activities for an active pelvis orthosis using two IMUs integrated into it. The designed on-board intent recognition system with a BPNN based algorithm realizes distinguish among six locomotion modes including standing, level ground walking, ramp ascending, ramp descending, stair ascending and stair descending, and deliver the recognition results for future control strategies. Experiments are conducted on one healthy subject including on-board training and online recognition parts. The overall recognition accuracy is 97.79% with the cost time of one recognition decision is about 0.9ms, which is sufficient short compared with the sample interval of 10ms. The experimental results validate the great performance of the proposed real-time on-board training and recognition method for future control of the lower-limb exoskeletons assisting in various locomotion modes or terrains.

Attachments: Video Attachment
Keywords: Novel mechanism design, Novel actuation mechanisms, System integration
Abstract: A three-finger under actuated robotic hand with dexterous force control and inherent compliance is developed and tested. A simplified biomimetic finger design is generated and applied, with mechanical intelligence principles carefully designed and embedded such that optimal trajectories for grabbing are naturally followed and the fingers can automatically conform to the goal object. A generalizable potential energy flow theory is then proposed to explain the mechanism behind the mechanical intelligence. The theory is also supported by experimental results. Quasi-direct drive actuators were developed to actuate the robotic hand with proprioceptive force sensing and inherent compliance. The hand performs delicate force controlled manipulation with a simple compliance controller implemented.

Attachments: Video Attachment
Keywords: Learning from demonstration, Skill modelling, Grasping and Manipulation
Abstract: This paper proposes a novel force control design method, and it is used to assemble a ring-shaped elastic part to a cylinder’s outer groove. To assemble a ring-shaped elastic part, forces acting on an elastic part should be made as small as possible. To cope with this problem, we propose a novel method in which the force control strategy itself is automatically determined based on the human characteristics while the parameters of the controller are determined by using a numerical optimization. First, the position data and the force data while a human demonstrates the assembly are measured. From the measured data, two control methods are derived by using the normalized cross-correlation (NCC). Then, we optimize the parameters included in the obtained controller by using the downhill simplex method. The objective function of optimization is the peak force during the assembly. We confirmed that the applied force is considerably reduced compared with conventional methods.

Attachments: Video Attachment
Keywords: Novel mechanism design, Grasping and Manipulation, Trajectory planning
Abstract: Grasp planning and motion synthesis for dexterous manipulation tasks are traditionally done given a pre-existing kinematic model for the robotic hand. In this paper, we introduce a framework for automatically designing hand topologies best suited for manipulation tasks given high level objectives as input. Our goal is to create a fast pipeline that automatically generates custom hands designed for specific manipulation tasks based on high level user input. Our framework comprises of a sequence of trajectory optimizations chained together to translate a sequence of objective poses into an optimized hand mechanism along with a physically feasible motion plan involving both the constructed hand and the object. We demonstrate the feasibility of this approach by synthesizing a series of hand designs optimized to perform specified in-hand manipulation tasks of varying difficulty.

Keywords: Prostheses & Ortheses, Exoskeletons and assistive devices, Grasping and Manipulation
Abstract: Myocontrol is the control of an assistive device via the interpretation of the subject’s intent using surface electromyography, and one paradigmatic instance of myocontrol is in upper-limb prosthetics applications. The reliability of this kind of control remains a key issue — effective and stable upper-limb myocontrol is one of the most interesting open problems in the field of human-robot interfaces and rehabilitation. In this work we focused on the myocontrol of a prosthetic hand while grasping: performing grasp actions only when, and exactly for the duration, the user desires, avoiding failures that can lead to frustrating or catastrophic results. One specific step to improve stability in the myocontrol of prosthetic hands is the possibility to automatically detect the occurrence of a failure. For this purpose, the availability of an automatic “oracle” able to accomplish this work enables the possibility of self-adaptation of the myocontrol system—e.g. via on-demand model updates for incremental learning. According to this view, we performed an experiment using a simplified but still realistic grasping protocol involving four able-bodied expert myocontrol users, and we extracted features from a state-of-the-art commercial prosthetic hand to automatically identify instability in the myocontrol. The results show that a standard classifier is able to detect failures with a mean balanced error rate of 15.98% over the subjects that took part in the experiments. Our results can also be potentially applied in non-medical applications such as, e.g., teleoperation using extra-light interfaces.

Keywords: Grasping and Manipulation, Novel actuation mechanisms, Grasp and motion planning
Abstract: Human fingers are highly dexterous due to the combination of multiple joints and degrees of freedom (DoFs). With the rise of soft material robots, many gripper prototypes have utilized soft robot technology. Nonetheless, it is still challenging to build a soft finger with a similar dexterity to human hand. Owing to the soft pneumatic actuator (SPA) fabrication flexibility, we have proposed a multiple DoFs soft pneumatic fingers design scheme that can mimic the motion patterns of human hand. Unlike the most SPA fingers where the motion patterns are fixed and limited, here our soft fingers can function like the human finger movements to a greater extent. The design is composed of three parts: (i) a SPA with two independent chambers as the main part of finger; (ii) a constraint layer made of unstretchable fabric which simulates tendon; (iii) a fiber-reinforced three-channel fluidic elastomer actuator (FEA). The abundance in DoFs expands the range of motions of the fingers and enables them to reach where human fingers can go. By controlling the internal pressure of the actuator, a variety of human finger motion patterns are achieved. Since each chamber's pressure can be individually controlled, the dexterous position control of the finger has been made possible. Fabrication process of the soft pneumatic fingers is presented, followed by the characterizations to provide a better understanding of their behaviors. In the last part of the article, a robotic hand composed of five fingers is fabricated for demonstration.

Attachments: Video Attachment
Keywords: Grasp and motion planning, Trajectory planning
Abstract: We consider the problem of grasping in clutter. While there have been motion planners developed to address this problem in recent years, these planners are mostly tailored for open-loop execution. Open-loop execution in this domain, however, is likely to fail, since it is not possible to model the dynamics of the multi-body multi-contact physical system with enough accuracy, neither is it reasonable to expect robots to know the exact physical properties of objects, such as frictional, inertial, and geometrical. Therefore, we propose an online re-planning approach for grasping through clutter. The main challenge is the long planning times this domain requires, which makes fast re-planning and fluent execution difficult to realize. In order to address this, we propose an easily parallelizable stochastic trajectory optimization based algorithm that generates a sequence of optimal controls. We show that by running this optimizer only for a small number of iterations, it is possible to perform real time re-planning cycles to achieve reactive manipulation under clutter and uncertainty.

Attachments: Video Attachment
Keywords: Grasp and motion planning, Grasping and Manipulation, Body balancing
Abstract: Autonomous behaviors in humanoid robots are generally implemented by considering the robot as two separate parts, using the lower body for locomotion and balancing, and the upper body for manipulation actions. This paper provides a unified framework for autonomous grasping with bipedal robots using a compliant whole-body controller. The grasping action is based on parametric grasp planning for unknown objects using shape primitives, which allows a generation of multiple grasp poses on the object. A reachability analysis is used to select the final grasp, and also for triggering a base repositioning behavior that locates the robot on a better position for grasping the desired object more confidently, considering all grasps and the uncertainty in reaching the desired position. The whole-body controller accounts for perturbations at any level and ensures a successful execution of the intended task. The approach is implemented in the humanoid robot TORO, and different experiments demonstrate its robustness and flexibility.

Attachments: Video Attachment
Keywords: Grasping and Manipulation, Industrial
Abstract: This paper presents a double jaw hand for industrial assembly. The hand comprises two orthogonal parallel grippers with different mechanisms. The inner gripper is made of a crank-slider mechanism which is compact and able to firmly hold objects like shafts. The outer gripper is made of a parallelogram that has large stroke to hold big objects like pulleys. The two grippers are connected by a prismatic joint along the hand's approaching vector. The hand is able to hold two objects and perform in-hand manipulation like pull-in (insertion) and push-out (ejection). This paper presents the detailed design and implementation of the hand, and demonstrates the advantages by performing experiments on two sets of peg-in-multi-hole assembly tasks as parts of the World Robot Challenge (WRC) 2018 using a bimanual robot.