Statistical Learning for Humanoid Robots

High performance and compliant robot control requires accurate dynamics models which cannot be obtained analytically for sufficiently complex robot systems. In such cases, machine learning offers a promising alternative for approximating the robot dynamics using measured data. This approach offers a natural framework to incorporate unknown nonlinearities as well as to continually adapt online for changes in the robot dynamics. However, the most accurate regression methods, e.g. Gaussian processes regression (GPR) and support vector regression (SVR), suffer from exceptional high computational complexity which prevents their usage for large numbers of samples or online learning to date. Inspired by locally linear regression techniques, we propose an approximation to the standard GPR using local Gaussian processes models inspired by. Due to reduced computational cost, local Gaussian processes (LGP) can be applied for larger sample-sizes and online learning. Comparisons with other nonparametric regressions, e.g. standard GPR, nu-SVR and locally weighted projection regression (LWPR), show that LGP has higher accuracy than LWPR and close to the performance of standard GPR and nu-SVR while being sufficiently fast for online learning.

The author takes a global rather than instantaneous look at the inverse kinematics of redundant manipulators. This approach is based on a manifold mapping reformulation of manipulator kinematics. While the kinematic problem has an infinite number of solutions for redundant manipulators, the infinity of solutions can be grouped into a finite and bounded set of disjoint continuous manifolds. Each of these manifolds, termed self-motion manifolds, physically corresponds to a distinct self-motion of the manipulator, and the number, geometry, and characterizations of the self-motion manifolds are investigated. >

Summary form only given, as follows. A feedforward neural network was used to solve the problem of inverse kinematics for the redundant robots. A learning algorithm was also developed for the training of the network. The convergence of the training process was guaranteed according to Liapunov's stability theory. Moreover, the speed of training can be increased by increasing a learning rate parameter. Simulation was done to illustrate the effectiveness of the proposed network. >

While recent advances in approaches for control of humanoid robot systems show highly promising results, consideration of fully integrated humanoid systems for solving complex tasks such as disaster response has only gained focus recently. In this work, a software framework for humanoid disaster response robots is introduced. It provides newcomers as well as experienced researchers in humanoid robotics a comprehensive system comprising open source packages for locomotion, manipulation, perception, world modeling, behavior control, and operator interaction. The system uses the Robot Operating System (ROS) as a middleware, which has emerged as a de-facto standard in robotics research in recent years. The described architecture and components allow for flexible interaction between operator(s) and robot from teleoperation to remotely supervised autonomous operation while considering bandwidth constraints. The components are self-­contained and can be used either in combination with others or standalone. They have been developed and evaluated during participation in the DARPA Robotics Challenge and their use for different tasks and parts of this competition are described.

Problem statement: Present researches focus to make humanoid robots more and more autonomous so they can assist human in daily works like taking care of children, aged or disabled persons. In such social activities, the contemporar y humanoid robots are expected to have human like morphology and gait. Studies on bipedal locomotion for humanoid robots are then part of the hottest topics in the field of robotic researches. Knowing the benefits of arm swinging for human gait, we propose in this study a new prototype of female humanoid robot morphology having the capabilities to swing arms during step walking. Approach: A new humanoid robot prototype had been introduced based on a human morphology corresponding to a woman whose weight is 70 kg and height is 1,73 m and using realistic gait parameters of a women. The female hu manoid robot prototype was composed of fifteen links associated to twenty-six degrees of freedom. Winter statistical model had been applied to determ ine all physical parameters corresponding to each link. Modeling the proposed humanoid robot implies first to establish the kinematic model basically founded on Euler's transformation matrix and then to set th e dynamic model computed using the Newton-Euler method. To show how the arms played an important role in bipedal gait, we had chosen to consider the whole body as two independent robotic systems: the upper body and the lower body. Results: Both three dimensional kinematic and dynamic models of the humanoid robot had been developed. The three dimensional humanoid robot was controlled via a feedback linearization control during the single su pport, impact and double support phases. The simulation results showed the arm swing during the step of walking. Conclusion: The humanoid robot proposed has a human like morphology and ensures the function of a step walking with arm swinging. The applied control laws have ensured to the robot desired performances during a step walking.

This paper presents a component-based framework of humanoid robot task coordination using Predicate Transition Nets (PrT Nets). Humanoid robots have a large number of degrees of freedom (DOF) and they are expected to generate human-like stable behaviors to finish missions. Their motions have to satisfy a set of constraints (balance, coordinated motions, collision free movements), which increases the complexity of the system - each movement may consist of several types of movements and tasks simultaneously. Task is defined as a sequence of movements and/or actions towards to completing a given mission. It is key to represent the task coordination precisely and correctly to program on humanoid robots' motions. To solve the challenge issue of multiple task coordination for a humanoid robot, we presented a framework that integrates Component Based Software Development (CBSD) with Predicate Transition Nets to analyze the correctness of task co ordinations. Each component's behavior can be represented by a Predicate Transition Net, and the constraints of the behaviors are described by a set of temporal logic formulae. To ensure the task co ordinations are met, a rewriting logic-based model checker is applied to verify the system against the constraints. This formal framework is general and can be used for other type humanoid robots. It allows the representation of motion plans and provides the flexibility of semantic analysis on the humanoid robotics systems.

Locally weighted learning (LWL) is a class of statistical learning techniques that provides useful representations and training algorithms for learning about complex phenomena during autonomous adaptive control of robotic systems. This paper introduces several LWL algorithms that have been tested successfully in real-time learning of complex robot tasks. We discuss two major classes of LWL, memory-based LWL and purely incremental LWL that does not need to remember any data explicitly. In contrast to the traditional beliefs that LWL methods cannot work well in high-dimensional spaces, we provide new algorithms that have been tested in up to 50 dimensional learning problems. The applicability of our LWL algorithms is demonstrated in various robot learning examples, including the learning of devil-sticking, pole-balancing of a humanoid robot arm, and inverse-dynamics learning for a seven degree of-freedom robot.

In this paper, a new method is proposed to solve the inverse kinematics problem of redundant manipulators. This method demonstrates superior performance on continuous motion by combining interval search genetic algorithm based on trajectory which we propose with parametric joint angle method. In this method, population continuity strategy is utilized to improve search speed and reduce evolutionary generation, interval search strategy is introduced to enhance the search ability and overcome the influence of singularity, and reference point strategy is used to avoid sudden changes of joint variables. By introducing those three strategies, this method is especially suitable for redundant manipulators that perform continuous motion. It can not only obtain solutions of inverse kinematics quickly, but also ensure the motion continuity of manipulator and accuracy of the end effector. Moreover, this algorithm can also perform multi-objective tasks by adjusting the fitness function. Finally, this algorithm is applied to an 8 degree of freedom tunnel shotcrete robot. Field experiments and data analysis show that the algorithm can solve the problem quickly in industrial field, and ensure the motion continuity and accuracy.

Robotic manipulators can execute multiple tasks precisely at the same time and, thus, the task-priority scheme plays an important role in implementing multiple tasks. Until now, several algorithms for task-priority have been used in solving the inverse kinematics for redundant manipulators. In this paper, through the comparative study of existing algorithms, we will propose a new method for task-priority manipulation in terms of two important criteria—algorithmic singularity and task error. This manipulation scheme will be applied to a planar three-link manipulator to demonstrate its effectiveness.

Presents our finding of a critical point that has not been reported thus far in the inverse kinematics of redundant manipulators under inequality constraints. The critical point, named algorithmic barrier was first encountered in the form of peculiar phenomena in our experiment with an 8-DOF robot, which we believe to have significance in the order of so-called algorithmic singularity. In addition, the paper deals with the characteristics of optimal solutions (COS) in resolving manipulator redundancy under inequality constraints. In order to analyze the COS, analytic functions of sufficient conditions and critical point conditions are derived. As a result, we find that COS is drastically affected by the introduction of inequality constraints. That is, COS under no inequality constraints is known to change only at algorithmic singularity while COS under inequality constraints turns out to change at semi-singularity and algorithmic barrier as well as algorithmic singularity. We prove the existence the critical points and present their analytical properties by using a planar 3-DOF manipulator.

Trial mountain climbing algorithm to solve the inverse kinematics problem of redundant manipulator is introduced, and a method of describing a numeral with a special numeration system is given to define the changed step of the trial mountain climbing algorithm. The results show that a likelihood solution can be found quickly in the infinite groups of likelihood solutions within the limited search times, and need not calculate the anti-trigonometric function and the inverse matrix. In addition, this algorithm has many good qualities such as concise algorithm, tiny computation, fast convergence velocity, good stability and extensive adaptability.

In this paper, the second order recurrent neural network is adopted to study the inverse kinematics problem of three degree-of-freedom planar redundant manipulators. The Simulation results show that the network can effectively solve the inverse kinematics problem of redundant manipulators, and it reaches to good precision of solution and solving speed.

Solving the inverse kinematics of redundant manipulators is difficult, because knowledge of the manipulators and their evaluation functions is required. To solve this problem, we propose a novel method of enabling a neural network model to learn the inverse kinematics. The method achieves learning independent of the structure of the evaluation function, by combining multiple neural network models. The method can obtain the neural network models of the inverse kinematics via an automatic calculation process using only training data, which consist of the postures, end-points, and evaluation values. In this paper, the algorithm used by the method and its background is explained, and the effectiveness of the method is validated by a numerical simulation.

In Indonesia research on Humanoid is limited to Kid-Size only, while for Teen-Size is still rare. This encourages teams and researchers to create a Teen-size humanoid called “FLoW”. To perform flexible movements like humans is not easy to do robot. Some of the problems that arise and become the focus of this research is to organize the movement of the robot in order to move, synchronize the movement in every parts of the humanoid robot, mathematical modeling and simulation of robot movement when standing. Taking into account the total kinematics of the humanoid robot, the movement of the robot can be overcome. With forwardinvers kinematics, researchers can determine a movement of the robot by controlling the motor part at a certain point. This section is an actuator of the robot. To do the modeling and simulation using D-H parameter and V-REP simulation software aid. Then for forward-invers kinematics can be implemented on the PID algorithm with the output of the speed on the motor that can form an angle on the motor to drive the robot. The expected result of this research is Humanoid Robot can move to follow the trajectory that has been determined. Making Humanoid Teen-Size robot is expected to increase research on Humanoid Teen-Size in Indonesia. In addition, this FLoW robot can be a supporting material and reference for the development of Humanoid Teen-Size next.

With the development of the concept of Industry 4.0, research relating to robots is being paid more and more attention, among which the humanoid robot is a very important research topic. The humanoid robot is a robot with a bipedal mechanism. Due to the physical mechanism, humanoid robots can maneuver more easily in complex terrains, such as going up and down the stairs. However, humanoid robots often fall from imbalance. Whether or not the robot can stand up on its own after a fall is a key research issue. However, the often used method of hand tuning to allow robots to stand on its own is very inefficient. In order to solve the above problems, this paper proposes an automatic learning system based on Particle Swarm Optimization (PSO). This system allows the robot to learn how to achieve the motion of rebalancing after a fall. To allow the robot to have the capability of object recognition, this paper also applies the Convolutional Neural Network (CNN) to let the robot perform image recognition and successfully distinguish between 10 types of objects. The effectiveness and feasibility of the motion learning algorithm and the CNN based image classification for vision system proposed in this paper has been confirmed in the experimental results.