Chyon Hae Kim

Physical control of the rotation center of an unsupported object rope turning by a humanoid robot

The paper describes a physical control method for rotating an object that has no support from below. Conventionally, physical correction of the rotational center of an unsupported object while maintaining rotation around an axis that is located outside the object is difficult. In particular, for the case in which the object is soft and pliable, control is more difficult because gravity can easily affect the shape of the object. As an example, we selected a task in which a robot turns a rope using an end effector. In order to accomplish this task, we propose a novel control method. To control the rotational center to a target position while fixing the shape by centrifugal force, we formulated a method whereby a compensator is added to a control rule that enables the trajectory of the end effector to converge to a limit-cycle within a fixed radius. We applied the proposed control method to a simplified rope simulator. The results of the simulation revealed that the end effector of the robot and the center of mass of the simplified rope converge to a limit-cycle attractor. This result indicates that the system applies the forces that stabilize the shape of the rope and the location of the rotational center. In addition, we applied the proposed method to a rope turning task performed by a humanoid robot. The robot was able to turn a rope with one fixed end and could collaborate with a human. Both tasks could be realized by the same control method.

A GPU parallel computing method for LPUSS

Abstract We discuss the effective implementation of parallel processing for linear prediction-based uniform state sampling (LPUSS). In previous work, we proposed LPUSS as an optimization algorithm for mechanical motions that assures high optimality of the solutions and computational efficiency. In parallel computation, LPUSS requires balanced memory allocation and managed processing timing. In this paper, we propose an effective parallel computing method that assures high optimality and calculation efficiency in parallel processing using GPU processor. We conducted two experiments to validate the proposed method. In the first experiment, we compared single-thread processing for LPUSS and the proposed parallel processing. As a result of this experiment, calculation speed of LPUSS was about 4–20 times faster than that with single-thread CPU. In the second experiment, we applied the proposed method to the optimization of sixtuple inverted pendulum. As a result, the proposed method optimized the motion within 40 minutes. According to our survey, there is no other optimization method that is applicable to higher than quadruple inverted pendulum models with standard constraints.

Online motion selection for semi-optimal stabilization using reverse-time tree

This paper presents a general method for creating an approximately optimal online stabilization system. An optimal stabilization system is an ideal online system that can calculate each optimal motion leading to a stable mechanical goal state depending on the current state. We propose a system that selects each semi-optimal motion according to the current state from a reverse-time tree. To create the reverse-time tree, we applied rapid semi-optimal motion planning method (RASMO) to a reverse-time search from a stable state. We also developed an online motion selection technique. To validate the proposed method, we simulated the stabilization of a double inverted pendulum. When we used an optimization criteria, time optimal, the system quickly stabilized the pendulums posture and velocity. When we used higher resolution RASMO, the time approached the optimal time. The general framework proposed here is applicable to a variety of machines.

Online exploratory behavior acquisition model based on reinforcement learning

Discernment behavior is an exploratory behavior that supports object feature extraction and is a fundamental tool used by robots to orient themselves, operate objects, and establish knowledge. The main contribution of this paper is to propose an active perception model and analyzes the acquired motion patterns. In this study, we propose an active perception model in which a robot autonomously learns discernment behavior by interacting with multiple objects in its environment. During such interactions, the robot receives reinforcement signals according to the cluster distance of the observed data. In other words, we use a reinforcement learning approach to reward the successful recognition of objects. We apply our proposed model to a mobile robot simulation to observe its effectiveness. Results show that our proposed model effectively established intelligent strategies based on the relationship between object features and the robot’s configuration. In addition, we perform our experiments using real mobile robots and observe the suitability of the observed learned behaviors.

Physical Control of the Rotation of a Flexible Object — Rope Turning with a Humanoid Robot

Rope turning tasks are useful to explore rhythmic physical human–robot interaction. However, in traditional studies, a robot was not able to turn a rope by itself, because simultaneous control of three factors, i. e., energy transmission, rotational axis and centrifugal force, is difficult when a robot rotates a flexible object such as a rope. In this paper, we propose a method to control these three factors simultaneously. We developed the method by adding a compensator to an attractor that attracts the end-effector of a robot to a uniform circular motion within a fixed radius. In a rope turning simulation, the end-effector of a robot and the center of mass of a simplified rope converged to uniform circular motions. In addition, we applied the proposed method to a rope turning task performed by a humanoid robot. The robot was able to turn a rope with one fixed end or in cooperation with a human.

Vehicle Dynamics Modeling Using FAD Learning

Highly precise vehicle dynamics modeling is indispensable for self-driving technology. We propose a model learning framework, which utilizes FAD (The abbreviation of the capital letters of free dynamics, actuator, and disturbance.) learning, motor babbling, and dynamics learning tree. In the proposed framework, modeling error was decreased compared with conventional neural network approach. Also, this framework is applicable to online learning. In experiments, FAD learning and dynamics learning tree decreased learning error. The dynamics of a simulated car was learned using motor babbling. The proposed framework is applicable to a variety of mechanical systems.

Online exploratory behavior acquisition of mobile robot based on reinforcement learning

In this study, we propose an online active perception system that autonomously acquires exploratory behaviors suitable for each embodiment of mobile robots using online learning. We especially focus on a type of exploratory behavior that extracts object features useful for robots orientation and object operation. The proposed system is composed of a classification system and a reinforcement learning system. While a robot is interacting with objects, the classification system classifies observed data and calculates reward values according to the cluster distance of the observed data. On the other hand, the reinforcement learning system acquires effective exploratory behaviors useful for the classification according to the reward. We validated the effectiveness of the system in a mobile robot simulation. Three different shaped objects were placed beside the robot one by one. In this learning, the robot learned different behaviors corresponding to each object. The result showed that the behaviors were the exploratory behaviors that distinguish the difference of corner angles of the objects.

FAD learning: Separate learning for three accelerations -learning for dynamics of boat through motor babbling

This paper addresses the modeling and measurement of a small boat. In some fishing tasks, anchorage is not applicable in order to capture shellfishes or fishes efficiently. Currently, many fishermen are manually stabilizing boats simultaneously with the capturing task. We propose a boat modeling method named FAD learning. In this method, the free dynamics acceleration and actuator acceleration of a boat are learned using the online learning of two dynamics learning trees (DLTs), which are developed by us. In order to measure the position, velocity, and acceleration, we developed an image processing method with an underwater camera. In the experiment, the motor babbling of a boat was performed on a water pool. The dynamical data from the boat was learned by DLTs. The effectiveness of the modeling was confirmed through the validation of the velocity that was predicted by DLTs.

Fully Automated Learning for Position and Contact Force of Manipulated Object with Wired Flexible Finger Joints

We discuss about the modeling technology in the object manipulation of the robot arm that is equipped with flexible finger joints. In recent years, flexible robot fingers are getting attention because of their handling capability and safety. However, the position and contact force of manipulated object take much non-linear uncertainty from the flexibility. In this paper, we propose the modeling framework of the position and contact force of the manipulated object. The proposed framework is an online learning method that is composed of motor babbling, dynamics learning tree (DLT), and \(\epsilon \)-greedy method. In the experiments, the effectiveness of DLT was compared with neural network (NN), the effectiveness of the proposed framework was validated using a drawing task of a humanoid robot that equipped with flexible finger joints. The proposed framework was able to realize a fully automated incremental-manipulation-learning.

Executing optimized throwing motion on robot arm with free joint

Graphical Abstract We address the throwing motion optimization for robot. In order to pursue the best throwing motion, we may need heuristics/intuition free methods. We propose a throwing method that is composed of rapid semi-optimal motion-planning and output zeroing method. So as to execute the optimized trajectories in real rigid body systems, we need some compensations for the noises around the optimized trajectories. We introduce a compensation method for the optimized throwing motions of a robot arm with a free joint. To validate the effectiveness of the proposed method, we conducted a throwing experiment using a two-link arm. As a result of the experiment, the robot arm threw a ball with 63.7 km/h, which was the best record through the past experiments of this arm.