Motion Primitives Research Articles

Robot Task-Constrained Optimization and Adaptation with Probabilistic Movement Primitives

Enabling a robot to learn skills from a human and adapt to different task scenarios will enable the use of robots in manufacturing to improve efficiency. Movement Primitives (MPs) are prominent tools for encoding skills. This paper investigates how to learn MPs from a small number of human demonstrations and adapt to different task constraints, including waypoints, joint limits, virtual walls, and obstacles. Probabilistic Movement Primitives (ProMPs) model movements with distributions, thus providing the robot with additional freedom for task execution. We provide the robot with three modes to move, with only one human demonstration required for each mode. We propose an improved via-point generalization method to generalize smooth trajectories with encoded ProMPs. In addition, we present an effective task-constrained optimization method that incorporates all task constraints analytically into a probabilistic framework. We separate ProMPs as Gaussians at each timestep and minimize Kullback–Leibler (KL) divergence, with a gradient ascent–descent algorithm performed to obtain optimized ProMPs. Given optimized ProMPs, we outline a unified robot movement adaptation method for extending from a single obstacle to multiple obstacles. We validated our approach with a 7-DOF Xarm robot using a series of movement adaptation experiments.

Comparing Skill Transfer Between Full Demonstrations and Segmented Sub-Tasks for Neural Dynamic Motion Primitives

Programming by demonstration has shown potential in reducing the technical barriers to teaching complex skills to robots. Dynamic motion primitives (DMPs) are an efficient method of learning trajectories from individual demonstrations using second-order dynamic equations. They can be expanded using neural networks to learn longer and more complex skills. However, the length and complexity of a skill may come with trade-offs in terms of accuracy, the time required by experts, and task flexibility. This paper compares neural DMPs that learn from a full demonstration to those that learn from simpler sub-tasks for a pouring scenario in a framework that requires few demonstrations. While both methods were successful in completing the task, we find that the models trained using sub-tasks are more accurate and have more task flexibility but can require a larger investment from the human expert.

Towards Open-Set NLP-Based Multi-Level Planning for Robotic Tasks

This paper outlines a conceptual design for a multi-level natural language-based planning system and describes a demonstrator. The main goal of the demonstrator is to serve as a proof-of-concept by accomplishing end-to-end execution in a real-world environment, and showing a novel way of interfacing an LLM-based planner with open-set semantic maps. The target use-case is executing sequences of tabletop pick-and-place operations using an industrial robot arm and RGB-D camera. The demonstrator processes unstructured user prompts, produces high-level action plans, queries a map for object positions and grasp poses using open-set semantics, then uses the resulting outputs to parametrize and execute a sequence of action primitives. In this paper, the overall system structure, high-level planning using language models, low-level planning through action and motion primitives, as well as the implementation of two different environment modeling schemes—2.5 or fully 3-dimensional—are described in detail. The impacts of quantizing image embeddings on object recall are assessed and high-level planner performance is evaluated using a small reference scene data set. We observe that, for the simple constrained test command data set, the high-level planner is able to achieve a total success rate of 96.40%, while the semantic maps exhibit maximum recall rates of 94.69% and 92.29% for the 2.5d and 3d versions, respectively.

Efficient Robot Manipulation via Reinforcement Learning with Dynamic Movement Primitives-Based Policy

Reinforcement learning (RL) that autonomously explores optimal control policies has become a crucial direction for developing intelligent robots while Dynamic Movement Primitives (DMPs) serve as a powerful tool for efficiently expressing robot trajectories. This article explores an efficient integration of RL and DMP to enhance the learning efficiency and control performance of reinforcement learning in robot manipulation tasks by focusing on the forms of control actions and their smoothness. A novel approach, DDPG-DMP, is proposed to address the efficiency and feasibility issues in the current RL approaches that employ DMP to generate control actions. The proposed method naturally integrates a DMP-based policy into the actor–critic framework of the traditional RL approach Deep Deterministic Policy Gradient (DDPG) and derives the corresponding update formulas to learn the networks that properly decide the parameters of DMPs. A novel inverse controller is further introduced to adaptively learn the translation from observed states into various robot control signals through DMPs, eliminating the requirement for human prior knowledge. Evaluated on five robot arm control benchmark tasks, DDPG-DMP demonstrates significant advantages in control performance, learning efficiency, and smoothness of robot actions compared to related baselines, highlighting its potential in complex robot control applications.

Multi-layer perceptron-particle swarm optimization: A lightweight optimization algorithm for the model predictive control local planner

The model predictive control trajectory planner is a popular and effective robot local motion planner. However, it is challenging to satisfy real-time requirements and implement them on embedded platforms due to their high complexity of solving and reliance on optimization solvers. This letter reports a lightweight and efficient two-stage solving algorithm for the model predictive control planner. Firstly, the general form of the model predictive control local planning problem was specified and simplified by the motion primitives. Then, a two-stage solving method of multi-layer perceptron pre-solving and particle swarm optimization re-optimizing is developed after splitting the cost function into two pieces. An multi-layer perceptron neural network was designed and trained offline to learn the solution of the model predictive control local planner without considering obstacles after selecting the inputs and outputs. Next, to accomplish obstacle avoidance, the particle swarm optimization algorithm re-optimizes the trajectory based on the outputs of the neural network. The experiment results demonstrate that the multi-layer perceptron-particle swarm optimization algorithm can quickly and accurately solve local planning problems, guiding robots to complete global paths with the same efficiency as expert solvers. The average solving time has been reduced by over 90%, enabling the robot to increase its control frequency or adopt higher-quality complex motion primitives. The multi-layer perceptron-particle swarm optimization algorithm can also be used for various robots and motion primitives, with a wide range of potential applications.

Hierarchical Trajectory Planning Based on Adaptive Motion Primitives and Bilevel Corridor

Hierarchical Trajectory Planning Based on Adaptive Motion Primitives and Bilevel Corridor

Imitation Learning of Compression Pattern in Robotic-Assisted Ultrasound Examination Using Kernelized Movement Primitives

Imitation Learning of Compression Pattern in Robotic-Assisted Ultrasound Examination Using Kernelized Movement Primitives

Imitation Learning and Teleoperation Shared Control With Unit Tangent Fuzzy Movement Primitives

Imitation Learning and Teleoperation Shared Control With Unit Tangent Fuzzy Movement Primitives

Humanoid Robot Path Planning Using Rapidly Explored Random Tree and Motion Primitives

Path planning is an essential part of the control system of any mobile robot. In this article the path planner for a humanoid robot is presented. The short description of an universal control framework and the Motion Generation System is also presented. Described path planner utilizes a limited number of motions called the Motion Primitives. They are generated by Motion Generation System. Four different algorithms, namely the: Informed RRT, Informed RRT with random bias, and RRT with A* likeheuristics were tested. For the last one the version with biased random function was also considered. All mentioned algorithms were evaluated considering three dif ferent scenarios. Obtained results are described and discussed.

An Approach to Realize Generalized Optimal Motion Primitives Using Physics Informed Neural Networks

Abstract Autonomous manipulation is a challenging problem in field robotics due to uncertainty in object properties, constraints, and coupling phenomenon with robot control systems. Humans learn motion primitives over time to effectively interact with the environment. We postulate that autonomous manipulation can be enabled by basic sets of motion primitives as well, but do not necessitate mimicking human motion primitives. This work presents an approach to generalized optimal motion primitives using physics-informed neural networks. Our simulated and experimental results demonstrate that optimality is notionally maintained where the mean maximum observed final position percent error was 0.564% and the average mean error for all the trajectories was 1.53%. These results indicate that notional generalization is attained using a physics-informed neural network approach that enables near optimal real-time adaptation of primitive motion profiles.

Trajectory Augmentation Method Based on Dynamic Movement Primitives

Trajectory Augmentation Method Based on Dynamic Movement Primitives

Real-Time Trajectory Planning and Effectiveness Analysis of Intercepting Large-Scale Invading UAV Swarms Based on Motion Primitives

Real-Time Trajectory Planning and Effectiveness Analysis of Intercepting Large-Scale Invading UAV Swarms Based on Motion Primitives

Self‐supervised vessel trajectory segmentation via learning spatio‐temporal semantics

AbstractThe study of vessel trajectories (VTs) holds significant benefits for marine route management and resource development. VT segmentation serves as a foundation for extracting vessel motion primitives and enables analysis of vessel manoeuvring habits and behavioural intentions. However, existing methods relying on predefined behaviour patterns face high labelling costs, which hinder accurate pattern recognition. This paper proposes a self‐supervised vessel trajectory segmentation method (SS‐VTS), which segments VTs based on their inherent spatio‐temporal semantics. SS‐VTS adaptively divides VTs into cells of optimal size. Then, it extracts split points on different semantic levels from the multi‐dimensional feature sequence of the VTs using self‐supervised learning. Finally, spatio‐temporal distance fusion module is performed on split points to determine change points and obtain VT segments with multiple semantics. Experiments on a real automatic identification system datasets show that SS‐VTS achieves state‐of‐the‐art segmentation results compared to seven baseline methods.

Learning Underwater Intervention Skills Based on Dynamic Movement Primitives

Improving the autonomy of underwater interventions by remotely operated vehicles (ROVs) can help mitigate the impact of communication delays on operational efficiency. Currently, underwater interventions for ROVs usually rely on real-time teleoperation or preprogramming by operators, which is not only time-consuming and increases the cognitive burden on operators but also requires extensive specialized programming. Instead, this paper uses the intuitive learning from demonstrations (LfD) approach that uses operator demonstrations as inputs and models the trajectory characteristics of the task through the dynamic movement primitive (DMP) approach for task reproduction as well as the generalization of knowledge to new environments. Unlike existing applications of DMP-based robot trajectory learning methods, we propose the underwater DMP (UDMP) method to address the problem that the complexity and stochasticity of underwater operational environments (e.g., current perturbations and floating operations) diminish the representativeness of the demonstrated trajectories. First, the Gaussian mixture model (GMM) and Gaussian mixture regression (GMR) are used for feature extraction of multiple demonstration trajectories to obtain typical trajectories as inputs to the DMP method. The UDMP method is more suitable for the LfD of underwater interventions than the method that directly learns the nonlinear terms of the DMP. In addition, we improve the commonly used homomorphic-based teleoperation mode to heteromorphic mode, which allows the operator to focus more on the end-operation task. Finally, the effectiveness of the developed method is verified by simulation experiments.

Motion primitives and sample-based trajectory planning for fixed-wing unmanned aircraft

Navigating fixed-wing unmanned aircraft presents significant challenges, primarily stemming from the intricate dynamics involved and the demand for rapid planning. In this paper, we introduce an efficient approach that concurrently addresses nonlinear control and trajectory planning challenges. To tackle the control problem, we introduce a computationally efficient motion primitive that effectively maps feasible targets to control inputs. As for trajectory planning, we propose an enhanced, rapidly exploring, random tree algorithm capable of generating high-quality trajectories with minimal exploration. Our primary motivation is to seamlessly integrate the introduced motion primitive with a sample-based trajectory-planning technique, providing a comprehensive solution to the intricate challenges posed by nonlinear control and trajectory planning for fixed-wing unmanned aircraft. Experimental results showcase our approach’s superiority, delivering enhanced control precision, faster planning speeds, and improved planning quality.

Kinematic and neuromuscular characterization of cognitive involvement in gait control in healthy young adults.

The signature of cognitive involvement in gait control has rarely been studied using both kinematic and neuromuscular features. The present study aimed to address this gap. Twenty-four healthy young adults walked on an instrumented treadmill in a virtual environment under two optic flow conditions: normal (NOF) and perturbed (POF, continuous mediolateral pseudorandom oscillations). Each condition was performed under single-task and dual-task conditions of increasing difficulty (1-, 2-, 3-back). Subjective mental workload (raw NASA-TLX), cognitive performance (mean reaction time and d-prime), kinematic (steadiness, variability and complexity in the mediolateral and anteroposterior directions) and neuromuscular (duration and variability of motor primitives) control of gait were assessed. The cognitive performance and the number and composition of motor modules were unaffected by simultaneous walking, regardless of the optic flow condition. Kinematic and neuromuscular variability was greater under POF compared to NOF conditions. Young adults sought to counteract POF by rapidly correcting task-relevant gait fluctuations. The depletion of cognitive resources through dual-tasking led to reduced kinematic and neuromuscular variability and this occurred to the same extent regardless of simultaneous working memory (WM) load. Increasing WM load led to a prioritization of gait control in the mediolateral direction over the anteroposterior direction. The impact of POF on kinematic variability (step velocity) was reduced when a cognitive task was performed simultaneously, but this phenomenon was no modulated by WM load. Collectively, these results shed important light on how young adults adjust the processes involved in goal-directed locomotion when exposed to varying levels of task and environmental constraints.

Human–robot collaborative handling of curtain walls using dynamic motion primitives and real-time human intention recognition

Human–robot collaboration fully leverages the strengths of both humans and robots, which is crucial for handling large, heavy objects at construction sites. To address the challenges of human–machine cooperation in handling large-scale, heavy objects — specifically building curtain walls — a human–robot collaboration system was designed based on the concept of “human–centered with machine support”. This system allows the handling of curtain walls according to different human intentions. First, a robot trajectory learning and generalization model based on dynamic motion primitives was developed. The operator’s motion intent was then characterized by their speed, force, and torque, with the force impulse introduced to define the operator’s intentions for acceleration and deceleration. Finally, a collaborative experiment was conducted on an experimental platform to validate the robot’s understanding of human handling intentions and to verify its ability to handle curtain wall. Collaboration between humans and robots ensured a smooth and labor-saving handling process.

Neuromuscular Control Strategies in Basketball Shooting: Distance-Dependent Analysis of Muscle Synergies.

Basketball victory relies on an athlete's skill to make precise shots at different distances. While extensive research has explored the kinematics and dynamics of different shooting distances, the specific neuromuscular control strategies involved remain elusive. This study aimed to compare the differences in muscle synergies during basketball shooting at different distances, offering insights into neuromuscular control strategies and guiding athletes' training. Ten skilled shooting right-handed male basketball players participated as subjects in this experiment. Electromyographic (EMG) data for full-phase shooting were acquired at short (3.2 m), middle (5.0 m), and long (6.8 m) distances. Non-negative matrix decomposition extracted muscle synergies (motor modules and motor primitives) during shooting. The results of this study show that all three distance shooting can be broken down into three synergies and that there were differences in the synergies between short and long distances, with differences in motor primitive 1 and motor primitive 2 at the phase of 45% - 59% (p < 0.001, t* = 4.418), and 78% - 88% (p < 0.01, t* = 4.579), respectively, and differences in the motor module 3 found in the differences in muscle weights for rectus femoris (RF) (p = 0.001, d = -2.094), and gastrocnemius lateral (GL) (p = 0.001, d = -2.083). Shooting distance doesn't affect the number of muscle synergies in basketball shooting but alters synergy patterns. During long distance shooting training, basketball players should place more emphasis on the timing and synergistic activation of upper and lower limbs, as well as core muscles.

Fruit flexible collecting trajectory planning based on manual skill imitation for grape harvesting robot

The flexible collection is vital for the intelligent grape-picking robot. In view of the reliable operation method obtained by long-term training of human prior skill, a fruit flexible collecting trajectory planning method based on manual skill imitation for grape harvesting robot is proposed. The method involves capturing manual teaching trajectory data using a motion capture system, preprocessing the data, extracting features from multiple teaching trajectories, and forming a probability distribution through Gaussian mixture model – Gaussian mixture regression (GMM-GMR). And combined with the key point of manual operation trajectory, the general trajectory generated by GMR is segmented and further imitated by kernelized movement primitives (KMP) to obtain the reference trajectory, respectively. An optimization method for hyperparameter adaptation KMP (O-KMP) was proposed to meet the trajectory fitting effect of multiple key points. Mean square error (MSE) was used to evaluate the deviation of the trajectory from the reference trajectory. The partial optimal trajectory is selected and integrated into a single trajectory. Two experiments were conducted to investigate imitation: For the same starting and ending points task, the MSE of the trajectory generated by O-KMP decreased by 15.274% compared to the original fixed hyperparameter KMP. For different placement tasks, the MSE of the trajectory generated by the O-KMP decreased by 7.296% compared to the original KMP. Finally, the robot arm of the actuator visually presents different task states for the compliant placement of fresh grapes.

Exploring motor skill acquisition in bimanual coordination: insights from navigating a novel maze task

In this study, we introduce a novel maze task designed to investigate naturalistic motor learning in bimanual coordination. We developed and validated an extended set of movement primitives tailored to capture the full spectrum of scenarios encountered in a maze game. Over a 3-day training period, we evaluated participants’ performance using these primitives and a custom-developed software, enabling precise quantification of performance. Our methodology integrated the primitives with in-depth kinematic analyses and thorough thumb pressure assessments, charting the trajectory of participants’ progression from novice to proficient stages. Results demonstrated consistent improvement in maze performance and significant adaptive changes in joint behaviors and strategic recalibrations in thumb pressure distribution. These findings highlight the central nervous system’s adaptability in orchestrating sophisticated motor strategies and the crucial role of tactile feedback in precision tasks. The maze platform and setup emerge as a valuable foundation for future experiments, providing a tool for the exploration of motor learning and coordination dynamics. This research underscores the complexity of bimanual motor learning in naturalistic environments, enhancing our understanding of skill acquisition and task efficiency while emphasizing the necessity for further exploration and deeper investigation into these adaptive mechanisms.