Incremental kinesthetic teaching of motion primitives using the motion refinement tube

We present an approach for kinesthetic teaching of motion primitives for a humanoid robot. The proposed teaching method starts with observational learning and applies iterative kinesthetic motion refinement using a forgetting factor. Kinesthetic teaching is supported by introducing the motion refinement tube, which represents an area of allowed motion refinement around the nominal trajectory. On the realtime control level, the kinesthetic teaching is handled by a customized impedance controller, which combines tracking performance with compliant physical interaction and allows to implement soft boundaries for the motion refinement. A novel method for continuous generation of motions from a hidden Markov model (HMM) representation of motion primitives is proposed, which incorporates time information for each state. The proposed methods were implemented and tested using DLR??s humanoid upper-body robot Justin.     

MAHRU-M: A mobile humanoid robot platform based on a dual-network control system and coordinated task execution

This paper introduces a mobile humanoid robot platform able to execute various services for humans in their everyday environments. For service in more intelligent and varied environments, the control system of a robot must operate efficiently to ensure a coordinated robot system. We enhanced the efficiency of the control system by developing a dual-network control system. The network system consists of two communication protocols: high-speed IEEE 1394, and a highly stable Controller Area Network (CAN). A service framework is also introduced for the coordinated task execution by a humanoid robot. To execute given tasks, various sub-systems of the robot were coordinated effectively by this system. Performance assessments of the presented framework and the proposed control system are experimentally conducted. MAHRU-M, as a platform for a mobile humanoid robot, recognizes the designated object. The object’s pose is calculated by performing model-based object tracking using a particle filter with back projection-based sampling. A unique approach is used to solve the human-like arm inverse kinematics, allowing the control system to generate smooth trajectories for each joint of the humanoid robot. A mean-shift algorithm using bilateral filtering is also used for real-time and robust object tracking. The results of the experiment show that a robot can execute its services efficiently in human workspaces such as an office or a home.     

Few-experiential learning system of robotic picking task with selective dual-arm grasping

Recently, robots are introduced to warehouses and factories for automation and are expected to execute dual-arm manipulation as human does and to manipulate large, heavy and unbalanced objects. We focus on target picking task in the cluttered environment and aim to realize a robot picking system which the robot selects and executes proper grasping motion from single-arm and dual-arm motion. In this paper, we propose a few-experiential learning-based target picking system with selective dual-arm grasping. In our system, a robot first learns grasping points and object semantic and instance label with automatically synthesized dataset. The robot then executes and collects grasp trial experiences in the real world and retrains the grasping point prediction model with the collected trial experiences. Finally, the robot evaluates candidate pairs of grasping object instance, strategy and points and selects to execute the optimal grasping motion. In the experiments, we evaluated our system by conducting target picking task experiments with a dual-arm humanoid robot Baxter in the cluttered environment as warehouse.     

Incremental motion learning through kinesthetic teachings and new motion production from learned motions by a humanoid robot

This work presents a new incremental motion learning algorithm through kinesthetic teachings and a new motion production algorithm by combining learned motions in a humanoid robot. The proposed algorithms are useful for improving the motions that a humanoid robot can produce. The learning algorithm consists of data encoding, time alignment, dimensional reduction, parameter estimation in the Gaussian mixture model (GMM) of motions, GMM refinement, and motion generation steps. The overall procedure is built to be incremental. No historic data memorization is required in any step, and model parameters are enough information to generate motions. The motion production algorithm allows a robot to extract new motions simply from learned motions without requiring teaching sessions. A series of experiments with a humanoid robot serves to validate the performance of the proposed algorithms.     

A new paradigm of humanoid robot motion programming based on touch interpretation

Most humanoid soccer robot teams design the basic movements of their robots, like walking and kicking, off-line and manually. Once these motions are considered satisfactory, they are stored in the robot’s memory and played according to a high level behavioral strategy. Much time is spent in the development of the movements, and despite the significant progress made in humanoid soccer robots, the interfaces employed for the development of motions are still quite primitive. In order to accelerate development, an intuitive instruction method is desired. We propose the development of robot motions through physical interaction. In this paper we propose a ”teaching by touching” approach; the human operator teaches a motion by directly touching the robot’s body parts like a dance instructor. Teaching by directly touching is intuitive for instructors. However, the robot needs to interpret the instructor’s intention since tactile communication can be ambiguous. This paper presents a method to learn the interpretation of the touch meaning and investigates, through experiments, a general (shared among different users) and intuitive touch manner.     

Real-time Unsupervised Segmentation of human whole-body motion and its application to humanoid robot acquisition of motion symbols

An interactive loop between motion recognition and motion generation is a fundamental mechanism for humans and humanoid robots. We have been developing an intelligent framework for motion recognition and generation based on symbolizing motion primitives. The motion primitives are encoded into Hidden Markov Models (HMMs), which we call “motion symbols”. However, to determine the motion primitives to use as training data for the HMMs, this framework requires a manual segmentation of human motions. Essentially, a humanoid robot is expected to participate in daily life and must learn many motion symbols to adapt to various situations. For this use, manual segmentation is cumbersome and impractical for humanoid robots. In this study, we propose a novel approach to segmentation, the Real-time Unsupervised Segmentation (RUS) method, which comprises three phases. In the first phase, short human movements are encoded into feature HMMs. Seamless human motion can be converted to a sequence of these feature HMMs. In the second phase, the causality between the feature HMMs is extracted. The causality data make it possible to predict movement from observation. In the third phase, movements having a large prediction uncertainty are designated as the boundaries of motion primitives. In this way, human whole-body motion can be segmented into a sequence of motion primitives. This paper also describes an application of RUS to AUtonomous Symbolization of motion primitives (AUS). Each derived motion primitive is classified into an HMM for a motion symbol, and parameters of the HMMs are optimized by using the motion primitives as training data in competitive learning. The HMMs are gradually optimized in such a way that the HMMs can abstract similar motion primitives. We tested the RUS and AUS frameworks on captured human whole-body motions and demonstrated the validity of the proposed framework.     

Acquisition of a symbolic manipulation task model by attention point analysis

As a way of automatic programming of robot behavior, a method for building a symbolic manipulation task model from a demonstration is proposed. The feature of this model is that it explicitly stores the information about the essential parts of a task, i.e. interaction between a hand and an environmental object, or interaction between a grasped object and a target object. Thus, even in different environments, this method reproduces robot motion as similar as possible to that of humans to complete the task while changing the motion during non-essential parts to adapt to the current environment. To automatically determine the essential parts, a method called attention point analysis is proposed; this method searches for the nature of a task using multiple sensors and estimates the parameters to represent the task. A humanoid robot is used to verify the reproduced robot motion based on the generated task model.     

Generating human-like soccer primitives from human data

Recently, interest in analysis and generation of human and human-like motion has increased in various areas. In robotics, in order to operate a humanoid robot, it is necessary to generate motions that have strictly dynamic consistency. Furthermore, human-like motion for robots will bring advantages such as energy optimization.This paper presents a mechanism to generate two human-like motions, walking and kicking, for a biped robot using a simple model based on observation and analysis of human motion. Our ultimate goal is to establish a design principle of a controller in order to achieve natural human-like motions. The approach presented here rests on the principle that in most biological motor learning scenarios some form of optimization with respect to a physical criterion is taking place. In a similar way, the equations of motion for the humanoid robot systems are formulated in such a way that the resulting optimization problems can be solved reliably and efficiently.The simulation results show that faster and more accurate searching can be achieved to generate an efficient human-like gait. Comparison is made with methods that do not include observation of human gait. The gait has been successfully used to control Robo-Erectus, a soccer-playing humanoid robot, which is one of the foremost leading soccer-playing humanoid robots in the RoboCup Humanoid League.     

Stepping over obstacles with humanoid robots

The wide potential applications of humanoid robots require that the robots can walk in complex environments and overcome various obstacles. To this end, we address the problem of humanoid robots stepping over obstacles in this paper. We focus on two aspects, which are feasibility analysis and motion planning. The former determines whether a robot can step over a given obstacle, and the latter discusses how to step over, if feasible, by planning appropriate motions for the robot. We systematically examine both of these aspects. In the feasibility analysis, using an optimization technique, we cast the problem into global optimization models with nonlinear constraints, including collision-free and balance constraints. The solutions to the optimization models yield answers to the possibility of stepping over obstacles under some assumptions. The presented approach for feasibility provides not only a priori knowledge and a database to implement stepping over obstacles, but also a tool to evaluate and compare the mobility of humanoid robots. In motion planning, we present an algorithm to generate suitable trajectories of the feet and the waist of the robot using heuristic methodology, based on the results of the feasibility analysis. We decompose the body motion of the robot into two parts, corresponding to the lower body and upper body of the robot, to meet the collision-free and balance constraints. This novel planning method is adaptive to obstacle sizes, and is, hence, oriented to autonomous stepping over by humanoid robots guided by vision or other range finders. Its effectiveness is verified by simulations and experiments on our humanoid platform HRP-2.     

Constrained Analytical Trajectory Filter for stabilizing humanoid robot motions

Mimicking human motion with a humanoid robot is essential for allowing humanoid robots to be used in service applications. Simply creating motions without considerations for balance and stability or directly copying motion from a human using motion capture and implementing it on a humanoid robot may not be successful because of the difference in physical properties between the human and the humanoid robot, which may cause instability and make it fall. Using the Zero Moment Point as the stability criteria, this work proposes a Constrained Analytical Trajectory Filter as part of an Analytical Motion Filter, which stabilizes a reference motion that can come from human motion capture data, kinematic synthesis, or animation software. The resulting solutions used in the Constrained Analytical Trajectory Filter provide insight into the complex interactions of motion and stability. The solutions were verified in simulation and with hardware, showing that the analytical filter can be successfully applied for stabilizing reference motions for humanoid robots which may be unstable otherwise.     

B-Theta*: an Efficient Online Coverage Algorithm for Autonomous Cleaning Robots

We propose a novel approach to deal with the online complete-coverage task of cleaning robots in unknown workspaces with arbitrarily-shaped obstacles. Our approach is based on the boustrophedon motions, the boundary-following motions, and the Theta* algorithm known as B-Theta*. Under control of B-Theta*, the robot performs a single boustrophedon motion to cover an unvisited region. While performing the boustrophedon motion, if the robot encounters an obstacle with a boundary that has not yet been covered, it switches to the boundary mode to cover portions along the obstacle boundary, and then continues the boustrophedon motion until it detects an ending point. To move to an unvisited region, the robot detects backtracking points based on its accumulated knowledge, and applies an intelligent backtracking mechanism thanks to the proposed Theta* for multi-goals in order to reach the next starting point. Complete coverage is achieved when no starting point exists for a new boustrophedon motion. Computer simulations and experiments on real workspaces show that our proposed B-Theta* is efficient for the complete-coverage task of cleaning robots.     

Physiological and subjective evaluation of a human-robot object hand-over task

In the context of task sharing between a robot companion and its human partners, the notions of safe and compliant hardware are not enough. It is necessary to guarantee ergonomic robot motions. Therefore, we have developed Human Aware Manipulation Planner (Sisbot et al., 2010), a motion planner specifically designed for human–robot object transfer by explicitly taking into account the legibility, the safety and the physical comfort of robot motions. The main objective of this research was to define precise subjective metrics to assess our planner when a human interacts with a robot in an object hand-over task. A second objective was to obtain quantitative data to evaluate the effect of this interaction. Given the short duration, the “relative ease” of the object hand-over task and its qualitative component, classical behavioral measures based on accuracy or reaction time were unsuitable to compare our gestures. In this perspective, we selected three measurements based on the galvanic skin conductance response, the deltoid muscle activity and the ocular activity. To test our assumptions and validate our planner, an experimental set-up involving Jido, a mobile manipulator robot, and a seated human was proposed. For the purpose of the experiment, we have defined three motions that combine different levels of legibility, safety and physical comfort values. After each robot gesture the participants were asked to rate them on a three dimensional subjective scale. It has appeared that the subjective data were in favor of our reference motion. Eventually the three motions elicited different physiological and ocular responses that could be used to partially discriminate them.     

Whole-body hierarchical motion and force control for humanoid robots

Robots acting in human environments usually need to perform multiple motion and force tasks while respecting a set of constraints. When a physical contact with the environment is established, the newly activated force task or contact constraint may interfere with other tasks. The objective of this paper is to provide a control framework that can achieve real-time control of humanoid robots performing both strict and non strict prioritized motion and force tasks. It is a torque-based quasi-static control framework, which handles a dynamically changing task hierarchy with simultaneous priority transitions as well as activation or deactivation of tasks. A quadratic programming problem is solved to maintain desired task hierarchies, subject to constraints. A generalized projector is used to quantitatively regulate how much a task can influence or be influenced by other tasks through the modulation of a priority matrix. By the smooth variations of the priority matrix, sudden hierarchy rearrangements can be avoided to reduce the risk of instability. The effectiveness of this approach is demonstrated on both a simulated and a real humanoid robot.     

Experience-based imitation using RNNPB

Robot imitation is a useful and promising alternative to robot programming. Robot imitation involves two crucial issues. The first is how a robot can imitate a human whose physical structure and properties differ greatly from its own. The second is how the robot can generate various motions from finite programmable patterns (generalization). This paper describes a novel approach to robot imitation based on its own physical experiences. We considered the target task of moving an object on a table. For imitation, we focused on an active sensing process in which the robot acquires the relation between the object's motion and its own arm motion. For generalization, we applied the RNNPB (recurrent neural network with parametric bias) model to enable recognition/generation of imitation motions. The robot associates the arm motion which reproduces the observed object's motion presented by a human operator. Experimental results proved the generalization capability of our method, which enables the robot to imitate not only motion it has experienced, but also unknown motion through nonlinear combination of the experienced motions.     

MODELING AND SIMULATION OF THE HUMANOID ROBOT HOAP-3 IN THE OPENHRP3 PLATFORM

The aim of this work is to model and simulate the humanoid robot HOAP-3 in the OpenHRP3 platform. Our purpose is to create a virtual model of the robot so that different motions and tasks can be tested in different environments. This will be the first step before testing the motion patterns in the real HOAP-3. We use the OpenHRP3 platform for the creation and validation of the robot model and tasks. The procedure followed to reach this goal is detailed in this article. In order to validate our experience, different walking motions are tested and the simulation results are compared with the experimental ones.     

Interaction rule learning with a human partner based on an imitation faculty with a simple visuo-motor mapping

Imitation has been receiving increasing attention from the viewpoint of not simply generating new motions but also the emergence of communication. This paper proposes a system for a humanoid who obtains new motions through learning the interaction rules with a human partner based on the assumption of the mirror system. First, a humanoid learns the correspondence between its own posture and the partner’s one on the ISOMAPs supposing that a human partner imitates the robot motions. Based on this correspondence, the robot can easily transfer the observed partner’s gestures to its own motion. Then, this correspondence enables a robot to acquire the new motion primitive for the interaction. Furthermore, through this process, the humanoid learns an interaction rule that control gesture turn-taking. The preliminary results and future issues are given.     

Realizing whole-body tactile interactions with a self-organizing,multi-modal artificial skin on a humanoid robot

In this paper, we present a new approach to realize whole-body tactile interactions with a self-organizing, multi-modal artificial skin on a humanoid robot. We, therefore, equipped the whole upper body of the humanoid HRP-2 with various patches of CellulARSkin – a modular artificial skin. In order to automatically handle a potentially high number of tactile sensor cells and motors units, the robot uses open-loop exploration motions, and distributed accelerometers in the artificial skin cells, to acquire its self-centered sensory-motor knowledge. This body self-knowledge is then utilized to transfer multi-modal tactile stimulations into reactive body motions. Tactile events provide feedback on changes of contact on the whole-body surface. We demonstrate the feasibility of our approach on a humanoid, here HRP-2, grasping large and unknown objects only via tactile feedback. Kinesthetically taught grasping trajectories, are reactively adapted to the size and stiffness of different test objects. Our paper contributes the first realization of a self-organizing tactile sensor-behavior mapping on a full-sized humanoid robot, enabling a position controlled robot to compliantly handle objects.     

Learning to Acquire Whole-Body Humanoid Center of Mass Movements to Achieve Dynamic Tasks

This paper presents a novel approach for acquiring dynamic whole-body movements on humanoid robots focused on learning a control policy for the center of mass (CoM). In our approach, we combine both a model-based CoM controller and a model-free reinforcement learning (RL) method to acquire dynamic whole-body movements in humanoid robots. (i) To cope with high dimensionality, we use a model-based CoM controller as a basic controller that derives joint angular velocities from the desired CoM velocity. The balancing issue can also be considered in the controller. (ii) The RL method is used to acquire a controller that generates the desired CoM velocity based on the current state. To demonstrate the effectiveness of our approach, we apply it to a ball-punching task on a simulated humanoid robot model. The acquired whole-body punching movement was also demonstrated on Fujitsu's Hoap-2 humanoid robot.     

仿人机器人复杂动态动作设计及相似性研究

提出了一种基于人体运动的考虑节奏相似性的仿人机器人复杂动态动作设计方法. 首先, 把人体的运动分割成基本动作段, 给出了运动学约束, 讨论了复杂动态动作的稳定性调节方法. 然后, 提出了考虑运动节奏的仿人机器人模仿人体动作的相似性函数, 并给出了满足运动学约束和动力学稳定性、具有高相似性的运动轨迹求解方法. 最后, 通过在仿人机器人 BHR-2 上进行中国功夫``刀术'实验验证了该方法的有效性.