Shuhei Ikemoto

CB2: A child robot with biomimetic body for cognitive developmental robotics

This paper presents a new research platform, CB2, a child robot with biomimetic body for cognitive developmental robotics developed by the Socially-Synergistic Intelligence (Hereafter, Socio-SI) group of JST ERATO Asada Project. The Socio-SI group has focused on the design principles of communicative and intelligent machines and human social development through building a humanoid robot that has physical and perceptual structures close to us, that enables safe and close interactions with humans. For this purpose, CB2 was designed, especially in order to establish and maintain a long-term social interaction between human and robot. The most significant features of CB2 are a whole-body soft skin (silicon surface with many tactile sensors underneath) and flexible joints (51 pneumatic actuators). The fundamental capabilities and the preliminary experiments are shown, and the future work is discussed.

Physical Human-Robot Interaction: Mutual Learning and Adaptation

Close physical interaction between robots and humans is a particularly challenging aspect of robot development. For successful interaction and cooperation, the robot must have the ability to adapt its behavior to the human counterpart. Based on our earlier work, we present and evaluate a computationally efficient machine learning algorithm that is well suited for such close-contact interaction scenarios. We show that this algorithm helps to improve the quality of the interaction between a robot and a human caregiver. To this end, we present two human-in-the-loop learning scenarios that are inspired by human parenting behavior, namely, an assisted standing-up task and an assisted walking task.

Humanlike shoulder complex for musculoskeletal robot arms

In recent years, musculoskeletal robots are being intensively studied to exploit the advantages of biological musculoskeletal systems for robot developments. In these robots, it is very important to assure engineering, biomechanical, and anatomical plausibility at the same time. However, these requirements are often in contradiction. Especially, in the humans shoulder complex, mimicking the glenohumeral joint and the scapulothoracic joint has been difficult because of the need to assure a wide range of movement and the joints stability at the same time. In this paper, we propose mechanical structures to realize the functions of the glenohumeral joints and scapulothoracic joint. These structures were used to develop a musculoskeletal robot arm driven by pneumatic artificial muscles. In addition, in order to verify the feasibility of the robot arm, we present a dynamic motion in which the robot arm throws a ball by a simple control strategy.

Anthropomorphic Muscular–Skeletal Robotic Upper Limb for Understanding Embodied Intelligence

In this paper, we describe an anthropomorphic muscular–skeletal robotic upper limb and focus on its soft interaction with the environment. Two experiments are conducted to demonstrate the ability of the system: object recognition by dynamic touch and adaptive door opening. The first experiment shows that the compliant robot is advantageous for categorizing an object by shaking and the second experiment shows that the human-comparable compliant robot can open a door without precise control. The robot is expected to have comparable anisotropic compliance to that of a human, which can be utilized for realization of human-like adaptive behavior.

Physical interaction learning: Behavior adaptation in cooperative human-robot tasks involving physical contact

In order for humans and robots to engage in direct physical interaction several requirements have to be met. Among others, robots need to be able to adapt their behavior in order to facilitate the interaction with a human partner. This can be achieved using machine learning techniques. However, most machine learning scenarios to-date do not address the question of how learning can be achieved for tightly coupled, physical touch interactions between the learning agent and a human partner. This paper presents an example for such human in-the-loop learning scenarios and proposes a computationally cheap learning algorithm for this purpose. The efficiency of this method is evaluated in an experiment, where human care givers help an android robot to stand up.

Real-time inverse dynamics learning for musculoskeletal robots based on echo state Gaussian process regression

A challenging topic in articulated robots is the control of redundantly many degrees of freedom with artificial muscles. Actuation with these devices is difficult to solve because of nonlinearities, delays and unknown parameters such as friction. Machine learning methods can be used to learn control of these systems, but are faced with the additional problem that the size of the search space prohibits full exploration in reasonable time. We propose a novel method that is able to learn control of redundant robot arms with artificial muscles online from scratch using only the position of the end effector, without using any joint positions, accelerations or an analytical model of the system or the environment. To learn in real time, we use the so called online “goal babbling” method to effectively reduce the search space, a recurrent neural network to represent the state of the robot arm, and novel online Gaussian processes for regression. With our approach, we achieve good performance on trajectory tracking tasks for the end effector of two very challenging systems: a simulated 6 DOF redundant arm with artificial muscles, and a 7 DOF robot arm with McKibben pneumatic artificial muscles. We also show that the combination of techniques we propose results in significantly improved performance over using the individual techniques alone.

Development of a tendon-driven robotic finger for an anthropomorphic robotic hand

Our paper proposes a tendon-driven robotic finger based on an anatomical model of a human finger and a suitable method for its analysis. Our study aims to realize an anthropomorphic robotic hand that has the same characteristics and dexterity as that of a human hand, and it also aims to identify the advantages of the human musculoskeletal structure for application to the design and control of robot manipulators. When designing an anthropomorphic robotic hand, several devices are required to apply the human finger structure to a tendon-driven robotic finger. Reasons for this include that one of the human finger muscles, namely, the lumbrical muscle, is situated between tendons, which is an unfavorable configuration for the tendon-driven mechanism. Second, unlike a standard pulley used in a tendon-driven mechanism, some moment arms of the human finger change nonlinearly according to the joint angle. In our robotic finger design, we address these difficulties by rearranging its tendons and develop a mechanism to change the moment arm. We also propose a method to analyze and control this robotic fingers coordinating joints using non-stretch branching tendons based on the human extensor mechanism with a virtual tendon Jacobian matrix and the advantage is that this constraint virtually reduces the degrees-of-freedom (DOF) of the mechanism. Further, we build a prototype to confirm its motion using this method. In addition, we show that the state with the reduced DOF can be lost by external forces acting on the mechanism, and this condition can be changed manually by adjusting the tendon forces. This makes it possible to control the virtual DOFs to satisfy the requirements of the task. Finally, we discuss the benefits from anthropomorphic structures including the tendon arrangement, which mimic the human lumbrical muscle, and the above mentioned mechanism with non-linear moment arms from the perspective that there are two states of DOFs. These insights may provide new perspectives in the design of robotic hands.

Analysis of physical human-robot interaction for motor learning with physical help

In this paper we investigate physical human-robot interaction (PHRI) as an important extension of traditional HRI research. The aim of this research is to develop a humanoid robot that can work in the same spaces as humans. We first propose a new control system that takes advantage of inherent joint flexibility. The control system is applied on a new humanoid robot called CB2. In order to clarify the difference between successful and unsuccesful interaction, we conduct an experiment where a human subject has to help the CB2 robot in its rising-up behavior. We also develop a new measure that reveals the difference between smooth and nonsmooth physical interactions. An analysis of the experimentpsilas data, based on the introduced measure, shows significant differences between experts and beginners in human-robot interaction. Consequently, we assume that this measure can be used in the evaluation method required for a motor learning system that uses physical help from a human helper.

Advantages of flexible musculoskeletal robot structure in sensory acquisition

Morphological computation is the concept for which a well-designed hardware can bear part of the computational cost required for robot’s control and perception. So far, many musculoskeletal robots have been developed by taking inspiration from human’s one and shown superior motion performances. The use of pneumatic artificial muscles (PAMs) has been the key to realize these high performance. Additionally, PAMs have the possibility of being used as sensors for environmental information because they are flexible and backdrivable. In this research, we focus on clarifying how PAMs can contribute to morphological computation of robots driven by these actuators. In particular, we propose an analysis method based on transfer entropy and apply this method to the experimental data acquired by a musculoskeletal robot that opens a door.

Shoulder complex linkage mechanism for humanlike musculoskeletal robot arms.

The shoulder complex in the human body consists of the scapula, clavicle, humerus, and thorax and bears the load imposed by arm movements while at the same time realizing a wide range of motions. To mimic and exploit its role, several musculoskeletal robot arms with shoulder complex mechanisms have been developed. However, although many research groups have tried to design the structures using links and joints that faithfully correspond to the bones and joints in the human shoulder complex, its function has not been successfully reproduced because biologically plausible designs seriously compromise engineering plausibility. In this paper, we propose a linkage mechanism that can reproduce complex three-dimensional scapulo movements and considers the trade-off between biological and engineering plausibilities. Subsequently, the design was validated by driving the mechanism using pneumatic artificial muscles (PAMs) placed similarly to muscles in humans. Further, we present experiments in which the robot was controlled by surface electromyographic signals from a human. We show that the proposed design, due to its kinematic similarity with human musculoskeletal systems, eases the conversion between the surface electromyogram signals and the PAMs control inputs.