Abstract:By learning Variable Impedance Control policy, robot assistants can intelligently adapt their manipulation compliance to ensure both safe interaction and proper task completion when operating in human-robot interaction environments. In this paper, we propose a DMP-based framework that learns and generalizes variable impedance manipulation skills from human demonstrations. This framework improves robots$'$ adaptability to environment changes(i.e. the weight and shape changes of grasping object at the robot end-effector) and inherits the efficiency of demonstration-variance-based stiffness estimation methods. Besides, with our stiffness estimation method, we generate not only translational stiffness profiles but also rotational stiffness profiles that are ignored or incomplete in most learning Variable Impedance Control papers. Real-world experiments on a 7 DoF redundant robot manipulator have been conducted to validate the effectiveness of our framework.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to enable the robot assistant to learn from human demonstrations and generalize variable - impedance manipulation skills to ensure both safe interaction and successful task completion in the human - robot interaction environment. Specifically, the paper proposes a framework based on Dynamic Movement Primitives (DMP), which can learn and generalize variable - impedance manipulation skills from human demonstrations, thereby improving the adaptability of the robot to environmental changes (such as changes in the weight and shape of objects grasped by the end - effector). ### Core Problems of the Paper 1. **Safety and Task Completion** - Robots need to dynamically adjust their operational compliance (i.e., impedance) according to task requirements and the environment to ensure safe interaction and complete the target task. 2. **Generalization Ability** - Although the traditional DMP model can generalize motion trajectories, it ignores the generalization of operational compliance. The method proposed in this paper not only learns operational compliance but also can generalize it to new scenarios. 3. **Rotational Stiffness Estimation** - Most existing methods mainly focus on the estimation of translational stiffness and ignore rotational stiffness. This paper solves this problem by introducing the quaternion logarithm mapping function and can estimate translational and rotational stiffness simultaneously. ### Overview of Solutions - **Extended DMP Model**: By extending the DMP model and combining it with the stiffness scheduling equation, the synchronous generalization of pose trajectories and stiffness profiles is achieved. - **Quaternion Mapping**: Use the quaternion logarithm mapping function to convert quaternions into decoupled three - dimensional tangent vectors, thereby estimating rotational stiffness. - **Stiffness Indicator Function**: Generate stiffness profiles according to the variance of the demonstration trajectory. Low - variance regions correspond to high stiffness, and high - variance regions correspond to low stiffness. ### Experimental Verification - Experiments were carried out using a 7 - degree - of - freedom redundant robot manipulator (Franka Emika Panda) to verify the effectiveness of the proposed framework. - The experimental results show that the robot can learn reasonable variable - impedance manipulation skills from human demonstrations and can adapt to changes in the weight and shape of grasped objects. ### Summary The DMP - based framework proposed in this paper can not only learn variable - impedance manipulation skills from human demonstrations but also generalize them to new scenarios, improving the adaptability of the robot to environmental changes. The experimental results show that this framework has a good effect in practical applications. ### Formula Display 1. **Quaternion Logarithm Mapping Function** \[ \hat{u}=\log(\hat{q}) = \begin{cases} \frac{\arccos(\hat{q}_w)}{\|\hat{q}_x,\hat{q}_y,\hat{q}_z\|}(\hat{q}_x,\hat{q}_y,\hat{q}_z),&(\hat{q}_x,\hat{q}_y,\hat{q}_z)\neq\vec{0}\\ (0,0,0),&\text{otherwise} \end{cases} \] 2. **Quaternion Exponential Mapping Function** \[ \hat{q}=\exp(\hat{u}) = \begin{cases} (\cos\|\hat{u}\|,\sin\|\hat{u}\|\cdot\frac{\hat{u}}{\|\hat{u}\|}),&\hat{u}\neq(0,0,0)\\ 1=(1,0,0,0),&\text{otherwise} \end{cases} \] 3. **Stiffness Indicator Function** \[ k_l(t)=a_l(d_l(t)-d_{\max,l})^2 + k_{\min,l} \] where \[ a_l=\frac{k_{\max,l}-k_{\min,l}}{(d_{\min,l}-d_{\max,l})^2}>0 \] Through these formulas and methods.

Learning and Generalizing Variable Impedance Manipulation Skills from Human Demonstrations

Learning Variable Impedance Skills from Demonstrations with Passivity Guarantee

A neural network based framework for variable impedance skills learning from demonstrations

COMBINATION OF AFFINE DEFORMATION AND DYNAMIC MOVEMENT PRIMITIVE IN LEARNING HUMAN MOTION FOR REDUNDANT MANIPULATOR

Robot Variable Impedance Control and Generalizing from Human–Robot Interaction Demonstrations

Human-Robot Variable Impedance Skills Transfer Learning Based on Dynamic Movement Primitives

A DMP-based Framework for Efficiently Generating Complete Stiffness Profiles of Human-like Variable Impedance Skill from Demonstrations

Robot Variable Impedance Skill Transfer and Learning Framework Based on a Simplified Human Arm Impedance Model

Variable Impedance Learning from Human Demonstrations

Effective Learning and Online Modulation for Robotic Variable Impedance Skills.

An Extended DMP Framework for Robot Learning and Improving Variable Stiffness Manipulation

Learning Compliant Manipulation Tasks from Force Demonstrations

Learning Human Compliant Behavior from Demonstration for Force-Based Robot Manipulation

Efficient Learning Variable Impedance Control for Industrial Robots

Human-Robot Interaction and Demonstration Learning Mode Based on Electromyogram Signal and Variable Impedance Control

Deep Model Predictive Variable Impedance Control

A Unified Parametric Representation for Robotic Compliant Skills with Adaptation of Impedance and Force

Simultaneously Learning of Motion, Stiffness, and Force from Human Demonstration Based on Riemannian DMP and QP Optimization

Dynamic Skill Learning from Human Demonstration Based on the Human Arm Stiffness Estimation Model and Riemannian DMP

Variable impedance control on contact-rich manipulation of a collaborative industrial mobile manipulator: An imitation learning approach

Robot Learning from Human Demonstration of Peg-In-Hole Task