Motion Primitives Research Articles

Learning compliant robotic movements based on biomimetic motor adaptation

It is one of the great challenges for a robot to learn compliant movements in interaction tasks. The robot can easily acquire motion skills from a human tutor by kinematics demonstration, however, this becomes much more difficult when it comes to the compliant skills. This paper aims to provide a possible solution to address this problem by proposing a two-stage approach. In the first stage, the human tutor demonstrates the robot how to perform a task, during which only motion trajectories are recorded without the involvement of force sensing. A dynamical movement primitives (DMPs) model which can generate human-like motion is then used to encode the kinematics data. In the second stage, a biomimetic controller, which is inspired by the neuroscience findings in human motor learning, is employed to obtain the desired robotic compliant behaviors by online adapting the impedance profiles and the feedforward torques simultaneously. Several tests are conducted to validate the effectiveness of the proposed approach.

Learning peg-in-hole assembly using Cartesian DMPs with feedback mechanism

Purpose This paper aims to enable the robot to obtain human-like compliant manipulation skills for the peg-in-hole (PiH) assembly task by learning from demonstration. Design/methodology/approach A modified dynamic movement primitives (DMPs) model with a novel hybrid force/position feedback in Cartesian space for the robotic PiH problem is proposed by learning from demonstration. To ensure a compliant interaction during the PiH insertion process, a Cartesian impedance control approach is used to track the trajectory generated by the modified DMPs. Findings The modified DMPs allow the robot to imitate the trajectory of demonstration efficiently and to generate a smoother trajectory. By taking advantage of force feedback, the robot shows compliant behavior and could adjust its pose actively to avoid a jam. This feedback mechanism significantly improves the dynamic performance of the interactive process. Both the simulation and the PiH experimental results show the feasibility and effectiveness of the proposed model. Originality/value The trajectory and the compliant manipulation skill of the human operator can be learned simultaneously by the new model. This method adopted a modified DMPs model in Cartesian space to generate a trajectory with a lower speed at the beginning of the motion, which can reduce the magnitude of the contact force.

Development of lower-limb rehabilitation exercises using 3-PRS Parallel Robot and Dynamic Movement Primitives

The design of rehabilitation exercises applied to sprained ankles requires extreme caution, regarding the trajectories and the speed of the movements that will affect the patient. This paper presents a technique that allows a 3-PRS parallel robot to control such exercises, consisting of dorsi/plantar flexion and inversion/eversion ankle movements. The work includes a position control scheme for the parallel robot in order to follow a reference trajectory for each limb with the possibility of stopping the exercise in mid-execution without control loss. This stop may be motivated by the forces that the robot applies to the patient, acting like an alarm mechanism. The procedure introduced here is based on Dynamic Movement Primitives (DMPs).

Pauses Provide Effective Control for an Underactuated Oscillating Swimming Robot

We describe motion primitives and closed-loop control for a unique low-cost single-motor oscillating aquatic system: the Modboat. The Modboat is driven by the conservation of angular momentum, which is used to actuate two passive flippers in a sequential paddling motion for propulsion and steering. We propose a discrete description of the motion of the system, which oscillates around desired trajectories, and propose two motion primitives — one frequency based and one pause-based — with associated closed-loop controllers. Testing is performed to evaluate each motion primitive, the merits of each are presented, and the pause-based primitive is shown to be significantly superior. Finally, waypoint following is implemented using both primitives and shown to be significantly more successful using the pause-based motion primitive.

Ori-Pixel, a Multi-DoFs Origami Pixel for Modular Reconfigurable Surfaces

Reconfigurable surfaces using robotic pixels can produce discretize 3D contours, shapes and motion patterns. In existing devices, the rendering resolution and richness of shapes and patterns are limited by the use of linear pixels that only produce single DoF (only change their extrusion/ height). Challenges in designing compact yet modular mechanisms and scaling their manufacturing have limited the development of more complex but versatile multi-DoFs pixels. Utilizing an origami approach, we have designed, prototyped and controlled the Ori-Pixel , a 3-DoFs parallel robot that can control its height, pitch and roll angles. These additional DoFs allow rendering a greater variety of shapes and motion primitives, replicating curved contours more accurately, and manipulating objects smaller than the actual pixel size. The origami folding joint design, multi-layer manufacturing and control of the Ori-Pixel are discussed in the letter. The workspace, stiffness, power consumption and communication rate of the Ori-pixel are experimentally characterized and confirmed with a prototype platform. We also report Ori-Pixels applications in two case studies: (i) an interactive flag consisting of a 5 × 5 pixel matrix that reacts to users’ gestures; (ii) a set of 33 Ori-Pixel to morph the surface of a concept car.

A Dexterous Soft Robotic Hand for Delicate In-Hand Manipulation

In this letter, we show that soft robotic hands provide a robust means of performing basic primitives of in-hand manipulation in the presence of uncertainty. We first discuss the design of a prototype hand with dexterous soft fingers capable of moving objects within the hand using several basic motion primitives. We then empirically validate the ability of the hand to perform the desired object motion primitives while still maintaining strong grasping capabilities. Based on these primitives, we examine a simple, heuristic finger gait which enables continuous object rotation for a wide variety of object shapes and sizes. Finally, we demonstrate the utility of our dexterous soft robotic hand in three real-world cases: unscrewing the cap of a jar, orienting food items for packaging, and gravity compensation during grasping. Overall, we show that even for complex tasks such as in-hand manipulation, soft robots can perform robustly without the need for local sensing or complex control.

Lower complexity of motor primitives ensures robust control of high-speed human locomotion

Walking and running are mechanically and energetically different locomotion modes. For selecting one or another, speed is a parameter of paramount importance. Yet, both are likely controlled by similar low-dimensional neuronal networks that reflect in patterned muscle activations called muscle synergies. Here, we challenged human locomotion by having our participants walk and run at a very broad spectrum of submaximal and maximal speeds. The synergistic activations of lower limb locomotor muscles were obtained through decomposition of electromyographic data via non-negative matrix factorization. We analyzed the duration and complexity (via fractal analysis) over time of motor primitives, the temporal components of muscle synergies. We found that the motor control of high-speed locomotion was so challenging that the neuromotor system was forced to produce wider and less complex muscle activation patterns. The motor modules, or time-independent coefficients, were redistributed as locomotion speed changed. These outcomes show that humans cope with the challenges of high-speed locomotion by adapting the neuromotor dynamics through a set of strategies that allow for efficient creation and control of locomotion.

A Task-Learning Strategy for Robotic Assembly Tasks from Human Demonstrations.

In manufacturing, traditional task pre-programming methods limit the efficiency of human–robot skill transfer. This paper proposes a novel task-learning strategy, enabling robots to learn skills from human demonstrations flexibly and generalize skills under new task situations. Specifically, we establish a markerless vision capture system to acquire continuous human hand movements and develop a threshold-based heuristic segmentation algorithm to segment the complete movements into different movement primitives (MPs) which encode human hand movements with task-oriented models. For movement primitive learning, we adopt a Gaussian mixture model and Gaussian mixture regression (GMM-GMR) to extract the optimal trajectory encapsulating sufficient human features and utilize dynamical movement primitives (DMPs) to learn for trajectory generalization. In addition, we propose an improved visuo-spatial skill learning (VSL) algorithm to learn goal configurations concerning spatial relationships between task-relevant objects. Only one multioperation demonstration is required for learning, and robots can generalize goal configurations under new task situations following the task execution order from demonstration. A series of peg-in-hole experiments demonstrate that the proposed task-learning strategy can obtain exact pick-and-place points and generate smooth human-like trajectories, verifying the effectiveness of the proposed strategy.

Skill Learning Strategy Based on Dynamic Motion Primitives for Human–Robot Cooperative Manipulation

This article presents a skill learning-based hierarchical control strategy for human-robot cooperative manipulation, which constitutes a novel learning-control system. The high-level learning strategy aims to learn the motor skills from human demonstrations by fusion with dynamic motion primitives (DMPs) and the Gaussian mixture model (GMM). The lower level control strategy guarantees the compliance of the robot movement under human interaction using admittance control and integral barrier Lyapunov function (IBLF)-based adaptive neural controller. First, the robot learns the motor skills from observing the successful execution of tasks by a demonstrator through DMP-GMM methods. Then, the robot reproduces the complex skills and executes the interactive task by demonstrations. Finally, the effectiveness of the proposed learning-control strategy is demonstrated with experimental results. The results show that the developed hierarchical strategy has good performance in cooperation by learning and control that reacts compliantly to robot interaction with human subjects.

Goal-Conditioned Variational Autoencoder Trajectory Primitives with Continuous and Discrete Latent Codes

Imitation learning is an intuitive approach for teaching motion to robotic systems. Although previous studies have proposed various methods to model demonstrated movement primitives, one of the limitations of existing methods is that the shape of the trajectories are encoded in high dimensional space. The high dimensionality of the trajectory representation can be a bottleneck in the subsequent process such as planning a sequence of primitive motions. We address this problem by learning the latent space of the robot trajectory. If the latent variable of the trajectories can be learned, it can be used to tune the trajectory in an intuitive manner even when the user is not an expert. We propose a framework for modeling demonstrated trajectories with a neural network that learns the low-dimensional latent space. Our neural network structure is built on the variational autoencoder (VAE) with discrete and continuous latent variables. We extend the structure of the existing VAE to obtain the decoder that is conditioned on the goal position of the trajectory for generalization to different goal positions. Although the inference performed by VAE is not accurate, the positioning error at the generalized goal position can be reduced to less than 1~mm by incorporating the projection onto the solution space. To cope with requirement of the massive training data, we use a trajectory augmentation technique inspired by the data augmentation commonly used in the computer vision community. In the proposed framework, the latent variables that encodes the multiple types of trajectories are learned in an unsupervised manner, although existing methods usually require label information to model diverse behaviors. The learned decoder can be used as a motion planner in which the user can specify the goal position and the trajectory types by setting the latent variables.

Fast human motion transfer based on a meta network

The paper proposes a new approach to the task of unsupervised human motion style transfer which transforms the style of an unlabeled input human motion into the style of another unlabeled motion sequence and preserves the behavior of the primitive motion at the same time. For supervised stylistic motion transfer task, constructing a database including a large quantity of training sample pairs is expensive and tedious for artists. Therefore, we focus on unsupervised methods. However, recent works need to retrain a network for each new character with different styles, which is a time-consuming and resource-intensive process. To tackle this problem, we utilize a meta network which builds a direct-mapping from the style motion sequence to the transformation network, namely, producing a corresponding motion transformation network in the meta network by a feed-forward propagation. The model transfers multiple styles of motion clips to different motions and generates natural stylized motions in real-time without a re-training for every style. Besides, our model can transfer styles between two movements depending on different skeletons. We also explore important features of the transfer network manifold by interpolation between the hidden states from meta network. Experiments validate the flexibility and effectiveness of our data-driven model.

Fractal analysis of muscle activity patterns during locomotion: pitfalls and how to avoid them.

Time-dependent physiological data sets are often difficult to interpret objectively. Biosignals such as electromyogram, electroencephalogram, or single-neuron recordings can be interpreted using various linear and nonlinear methods. Each analysis technique aims at the explanation of different data features that might be visible or not to the naked eye. Here, we used linear decomposition based on machine learning to extract motor primitives (the time-dependent coefficients of muscle synergies) from the hindlimb electromyographic activity of mice during normal and mechanically perturbed locomotion. We set out to investigate the effects of calculation parameters and data quality on two nonlinear metrics derived from fractal analysis: the Higuchi's fractal dimension (HFD) and the Hurst exponent (H). Both HFD and H proved to be exceptionally sensitive to changes in motor primitives induced by external perturbations to locomotion. We discuss the potential pitfalls that might arise from fractal analysis by using examples based on surrogate data. We conclude giving some simple, data-driven suggestions to reduce the chance of misinterpretations when metrics such as HFD and H are applied to any biological signal containing elements of periodicity.NEW & NOTEWORTHY Despite the lack of consensus on how to perform fractal analysis of physiological time series, many studies rely on this technique. Here, we shed light on the potential pitfalls of using the Higuchi's fractal dimension and the Hurst exponent. We expose and suggest how to solve the drawbacks of such methods when applied to data from normal and perturbed locomotion by combining in vivo recordings and computational approaches.

Hopf Bifurcations in Complex Multiagent Activity: The Signature of Discrete to Rhythmic Behavioral Transitions.

Most human actions are composed of two fundamental movement types, discrete and rhythmic movements. These movement types, or primitives, are analogous to the two elemental behaviors of nonlinear dynamical systems, namely, fixed-point and limit cycle behavior, respectively. Furthermore, there is now a growing body of research demonstrating how various human actions and behaviors can be effectively modeled and understood using a small set of low-dimensional, fixed-point and limit cycle dynamical systems (differential equations). Here, we provide an overview of these dynamical motor primitives and detail recent research demonstrating how these dynamical primitives can be used to model the task dynamics of complex multiagent behavior. More specifically, we review how a task-dynamic model of multiagent shepherding behavior, composed of rudimentary fixed-point and limit cycle dynamical primitives, can not only effectively model the behavior of cooperating human co-actors, but also reveals how the discovery and intentional use of optimal behavioral coordination during task learning is marked by a spontaneous, self-organized transition between fixed-point and limit cycle dynamics (i.e., via a Hopf bifurcation).

TossingBot: Learning to Throw Arbitrary Objects With Residual Physics

We investigate whether a robot arm can learn to pick and throw arbitrary rigid objects into selected boxes quickly and accurately. Throwing has the potential to increase the physical reachability and picking speed of a robot arm. However, precisely throwing arbitrary objects in unstructured settings presents many challenges: from acquiring objects in grasps suitable for reliable throwing, to handling varying object-centric properties (e.g., mass distribution, friction, shape) and complex aerodynamics. In this work, we propose an end-to-end formulation that jointly learns to infer control parameters for grasping and throwing motion primitives from visual observations (RGB-D images of arbitrary objects in a bin) through trial and error. Within this formulation, we investigate the synergies between grasping and throwing (i.e., learning grasps that enable more accurate throws) and between simulation and deep learning (i.e., using deep networks to predict residuals on top of control parameters predicted by a physics simulator). The resulting system, TossingBot, is able to grasp and successfully throw arbitrary objects into boxes located outside its maximum reach range at 500+ mean picks per hour (600+ grasps per hour with 85% throwing accuracy); and generalizes to new objects and target locations.

Vector Field Guided RRT* Based on Motion Primitives for Quadrotor Kinodynamic Planning

The intelligent drone is the key device of the future Internet of Drone, and its safe and robust flight in complicated environments still faces challenges. In this paper, we present a sampling-based kinodynamic planning algorithm for quadrotors, which plans a dynamically feasible trajectory in a complicated environment. We have designed a method to constrain the sampling state by using the vector field to construct a cone in the sampling stage of RRT*, so that the generated trajectory is connected as smoothly as possible to other states in the reachable set. The motion primitives are then generated by solving an optimal control problem and an explicit solution of the optimal duration for the motion primitives is given to optimally connect any pair of states. In addition, we have tried a new method to determine the neighbor radius for the non-Euclidean metrics of this paper. Finally, the planned trajectory is applied to the simulated quadrotor, which verifies the dynamic feasibility of the trajectory. Simulation results show that compared with the existing kinodynamic RRT* under the same number of iterations, the proposed algorithm explores more states with a shorter execution time and generates a smoother trajectory.

Learning Gait Models With Varying Walking Speeds

Lower-limb exoskeletons can reduce the therapist's burden and quantify repetitive gait training for patients with gait impairments. For patient's gait training, different walking speeds are required at different rehabilitation stages. However, due to the uniqueness of gait patterns, it is challenging for lower limb exoskeletons to generate individualized gait patterns for patients with different anthropometric parameters. This letter proposed learning-based gait models to learn and reconstruct gait patterns from healthy subject's gait database, including the Gait Parameters Model (GPM) and the Gait Trajectory Model (GTM). The GPM employs Neural Networks to predict gait parameters with a given desired walking speed and the anthropometric parameters of the subject. The GTM utilizes Kernelized Movement Primitives (KMP) to reconstruct gait patterns with the predicted gait parameters. The proposed approach has been tested on a lower limb exoskeleton named AIDER. Experimental results indicate that the reconstructed gait patterns are very similar to the subject's actual gait patterns for varying walking speeds.

Patterns of Whisker Movement Activity in the States of Sleep and Waking in Neonatal Rats

The aims of the present work was to describe primitive whisker movements in neonatal Wistar rats (postnatal days 4–7) in the states of sleep and waking and to study correlations between whisker movements and body movements. Whisker movements were recorded by high-speed video recording and the animals’ behavioral state was determined by recording electromyographic activity of the cervical muscles and simultaneous recording of limb movements using piezo elements. During active sleep, movements consisted mostly of short-term whisker movements, and a significant proportion of these movements arose simultaneously with short-term bursts of activity in the cervical muscles and transient myoclonic twitches of the animal’s body. During periods of waking, movements were mostly complex long-lasting whisker movements, which were accompanied by long-term increases in cervical muscle tone and complex body and limb movements. Both types of whisker movement occurred mainly in the rostral and caudal directions, and the amplitude of movements during waking was only slightly greater than the amplitude of whisker movements during sleep. The data obtained here provide evidence that the temporal organization of whisker movement activity in neonatal rats is significantly different in the states of sleep and waking and that whisker movements in both behavioral states showed strong correlations with the animal’s body movements. It is suggested that generalized primitive whisker and body movements in neonatal rats are due to the activity of a single central pattern generator both in the state of active sleep and in the state of waking.

A Framework for Recognition and Prediction of Human Motions in Human-Robot Collaboration Using Probabilistic Motion Models

This letter presents a framework for recognition and prediction of ongoing human motions. The predictions generated by this framework could be used in a controller for a robotic device, enabling the emergence of intuitive and predictable interactions between humans and a robotic collaborator. The framework includes motion onset detection, phase speed estimation, intent estimation and conditioning. For recognition and prediction of a motion, the framework makes use of a motion model database. This database contains several motion models learned using the probabilistic Principal Component Analysis (PPCA) method. The proposed framework is evaluated with joint angle trajectories of eight subjects performing squatting, stooping and lifting tasks. The motion onset and phase speed estimation modules are first evaluated separately. Next, an evaluation of the full framework provides more insight in the current challenges regarding motion prediction. A brief comparison between PPCA and the Probabilistic Movement Primitives (ProMP) method for learning motion models is made based on the influence of both methodologies on the performance of the framework. Both PPCA and ProMP motion models are able to predict motions over a short time horizon but struggle to predict motions over a longer horizon.

Task motion planning for anthropomorphic arms based on human arm movement primitives

PurposeThis paper aims to introduce the human arm movement primitive (HAMP) to express and plan the motions of anthropomorphic arms. The task planning method is established for the minimum task cost and a novel human-like motion planning method based on the HAMPs is proposed to help humans better understand and plan the motions of anthropomorphic arms.Design/methodology/approachThe HAMPs are extracted based on the structure and motion expression of the human arm. A method to slice the complex tasks into simple subtasks and sort subtasks is proposed. Then, a novel human-like motion planning method is built through the selection, sequencing and quantification of HAMPs. Finally, the HAMPs are mapped to the traditional joint angles of a robot by an analytical inverse kinematics method to control the anthropomorphic arms.FindingsFor the exploration of the motion laws of the human arm, the human arm motion capture experiments on 12 subjects are performed. The results show that the motion laws of human arm are reflected in the selection, sequencing and quantification of HAMPs. These motion laws can facilitate the human-like motion planning of anthropomorphic arms.Originality/valueThis study presents the HAMPs and a method for selecting, sequencing and quantifying them in human-like style, which leads to a new motion planning method for the anthropomorphic arms. A similar methodology is suitable for robots with anthropomorphic arms such as service robots, upper extremity exoskeleton robots and humanoid robots.

Learning Sequential Force Interaction Skills

Learning skills from kinesthetic demonstrations is a promising way of minimizing the gap between human manipulation abilities and those of robots. We propose an approach to learn sequential force interaction skills from such demonstrations. The demonstrations are decomposed into a set of movement primitives by inferring the underlying sequential structure of the task. The decomposition is based on a novel probability distribution which we call Directional Normal Distribution. The distribution allows infering the movement primitive’s composition, i.e., its coordinate frames, control variables and target coordinates from the demonstrations. In addition, it permits determining an appropriate number of movement primitives for a task via model selection. After finding the task’s composition, the system learns to sequence the resulting movement primitives in order to be able to reproduce the task on a real robot. We evaluate the approach on three different tasks, unscrewing a light bulb, box stacking and box flipping. All tasks are kinesthetically demonstrated and then reproduced on a Barrett WAM robot.