This paper proposes a real-time balance control technique that can be implemented to bipedal robots (exoskeletons, humanoids) whose ankle joints are powered via variable physical stiffness actuators. To achieve active balancing, an abstracted biped model, torsional spring-loaded flywheel, is utilized capture approximated angular momentum and stiffness, which of importance in postural balancing. In particular, this model enables us describe the mathematical relation between zero moment point...

Social demand for exoskeleton robots that physically assist humans has been increasing in various situations due to the demographic trends of aging populations. With robots, an assistive strategy is a key ingredient. Since interactions between users and are bidirectional, design problem complex challenging. In this paper, we explore data-driven learning approach designing strategies exoskeletons from user-robot physical interaction. We formulate as policy search exploit data-efficient...

In industrial assembly tasks, the in-hand pose of grasped objects needs to be known with high precision for subsequent manipulation tasks such as insertion. This problem (in-hand-pose estimation) has traditionally been addressed using visual recognition or tactile sensing. On one hand, while can provide efficient estimates, it tends suffer from low due noise, occlusions and calibration errors. other fingertip sensors precise complementary information, but their durability significantly...

Recent breakthroughs in wearable robots, such as exoskeleton robots with soft actuators and exosuits, have enabled the use of safe comfortable movement assistance. However, modeling identification methods for used yet to be sufficiently explored. In this study, we propose a novel approach obtaining accurate actuator models through design physical user–robot interactions which user applies external forces robot. To obtain an model from limited amount data acquired interaction, leverage active...

In this study, we present a novel control framework for assembly tasks with soft robot. Typically, existing hard robots require high frequency controllers and precise force/torque sensors tasks. The resulting robot system is complex, entailing large amounts of engineering maintenance. Physical softness allows the to interact environment easily. We expect perform without need sensors. However, specific data-driven approaches are needed deal complex models involving nonlinearity hysteresis. If...

This paper introduces a novel asynchronous adaptive brain machine interface (BMI), based on dry-wireless headset, to trigger the movement of lower limb exoskeleton robot by foot motor imagery. Specifically, it addresses two issues that are critical for development plug-and-play (BRI): setup-time and nonstationarity electroencephalogram (EEG). The former is solved headset reduces compared gel-based systems, removes nuisance cables. latter has been extensively studied in literature, leading...

In industrial assembly tasks, the position of an object grasped by robot has to be known with high precision in order insert or place it. real applications, this problem is commonly solved jigs that are specially produced for each part. However, they significantly limit flexibility and prohibitive when target parts change often, so a flexible method localize accuracy after grasping desired. To solve problem, we propose can estimate robot's hand sub-millimeter precision, improve its...

Physical softness has been proposed to absorb impacts when establishing contact with a robot or its workpiece, relax control requirements and improve performance in assembly insertion tasks. Previous work focused on special end effector solutions for isolated tasks, such as the peg-in-hole task. However, many tasks require precision of rigid robots, their would degrade simply adding compliance, it difficult take advantage physical real applications. A wrist that could switch between soft...

Deformable object manipulation has potential for a wide range of real-world applications, but is still largely unsolved due to the complex dynamics and difficulty state estimation. Learning-based approaches have recently accelerated progress, generally depend heavily on large simulated datasets, human demonstrations or both. In this study, we propose novel sample-efficient learning approach deformable linear (e.g. rope) that can be applied directly real world, without requiring any...

Designing an assistive strategy for exoskeletons is a key ingredient in movement assistance and rehabilitation. While several approaches have been explored, most studies are based on mechanical models of the human user, i.e., rigid-body dynamics or Center Mass (CoM)-Zero Moment Point (ZMP) inverted pendulum moECenter Massdel, only focus periodic movements with using oscillator models. On other hand, interactions between user robot often not considered explicitly because its difficulty...

Physically soft robots are promising for robotic assembly tasks as they allow stable contacts with the environment. In this study, we propose a novel learning system strategies. We formulate problem reinforcement task and design reward function from human demonstrations. Our key insight is that failed demonstrations can be used constraints to avoid behaviors. To end, developed teaching device which humans intuitively provide various Moreover, leverage Physically-Consistent Gaussian Mixture...

Recent studies suggest that reinforcement learning has great potential for generating assistive strategies in exoskeletons through physical interactions between a user and robot. Previous methods focused on task-specific strategy, where every single task (situation/context), the needs to interact with robot learn an appropriate strategy. Therefore, learned cannot be generalized new task. Since sampling cost is expensive such human-in-the-loop systems as exoskeletons, generalization must...

Transfer reinforcement learning (RL) aims at improving the efficiency of an agent by exploiting knowledge from other source agents trained on relevant tasks. However, it remains challenging to transfer between different environmental dynamics without having access environments. In this work, we explore a new challenge in RL, where only set policies collected under diverse unknown is available for target task efficiently. To address problem, proposed approach, MULTI-source POLicy AggRegation...

This paper introduces Elastic Decision Transformer (EDT), a significant advancement over the existing (DT) and its variants. Although DT purports to generate an optimal trajectory, empirical evidence suggests it struggles with trajectory stitching, process involving generation of or near-optimal from best parts set sub-optimal trajectories. The proposed EDT differentiates itself by facilitating stitching during action inference at test time, achieved adjusting history length maintained in...

This study aimed to anticipate fractures of fragile food during robotic manipulation. Anticipating allows a robot manipulate ingredients without irreversible failure. Food fracture models investigated in texture fields explain the properties objects well. However, they may not directly apply manipulation due variance physical even within same ingredient. To this end, we developed fracture-anticipation system with tactile sensing module and simple recurrent neural network. The key idea was...

Grinding materials into a fine powder is time-consuming task in material science that generally performed by hand, as current automated grinding machines might not be suitable for preparing small-sized samples. This study presents robotic system laboratory automation applications observe the powder's state to improve outcome. We developed soft jig consisting of off-the-shelf gel and 3D-printed parts, which can used with any robot arm perform grinding. The jig's physical softness allows safe...

This study focuses on a robotic powder weighing task used in laboratory automation. In this task, robot weighs certain amount of with milligram-level target mass using dispensing spoon. The complex dynamics the powder, variations materials being weighed, and need to balance conservative aggressive actions are significant challenges robotics field. Therefore, learning approaches critical for task. However, many interactions real-world environments require substantial efforts clean spread...

Practical industrial assembly scenarios often require robotic agents to adapt their skills unseen tasks quickly. While transfer reinforcement learning (RL) could enable such quick adaptation, much prior work has collect many samples from source environments learn target in a model-free fashion, which still lacks sample efficiency on practical level. In this work, we develop novel RL method named TRANSfer by Aggregating dynamics Models (TRANS-AM). TRANS-AM is based model-based (MBRL) for its...

We have successfully demonstrated a mechanochemical synthesis utilizing robotic powder grinding system capable of applying precisely controlled and constant mechanical force. Despite its signifi- cance, the application of...

Pneumatic Artificial Muscle (PAM) actuators have been used as exoskeletons because of their inherited compliance and high power-weight ratio. However, creating accurate models remains difficult mainly due to the issue; model can be changed by force applied user. Therefore, both user robot actions need considered for sufficient excitation PAMs that are equipped in exoskeleton robots, unlike typical rigid only sufficiently excited actions. In this paper, we propose a user-robot collaborative...

This paper presents a biologically-inspired real-time balance recovery control strategy that is applied to lower body exoskeleton with variable physical stiffness actuators at its ankle joints. For this purpose, torsional spring-loaded flywheel model presented encapsulate both approximated angular momentum and stiffness, which are crucial parameters in describing the postural balance. In particular, incorporation of compliance enables us provide three main contributions: i) A mathematical...