TrueRMA: Learning Fast and Smooth Robot Trajectories with Recursive Midpoint Adaptations in Cartesian Space

Imitation learning is a way to teach robots skills that are demonstrated by humans. Transfering skills between these different kinematic structures seems to be straightforward in Cartesian space. Because of the correspondence problem, however, the result will most likely not be identical. This is why refinement is required, for example, by policy search. Policy search in Cartesian space is prone to reachability problems when using conventional inverse kinematic solvers. We propose a configurable approximate inverse kinematic solver and show that it can accelerate the refinement process considerably. We also compare empirically refinement in Cartesian space and refinement in joint space.

In this article we present a technique that enhances an inverse kinematics (IK) solver such that when the results are applied to a computer character, we can generate a level of individualization tailored to both the character and the environment, e.g., a walking motion can become "stiffer" or can be turned into a limping motion. Since the technique is based on an IK solver, we also have the desirable effect of solving retargetting issues when mapping motion data between characters. As the individualization aspect of our technique is very tightly coupled with the inverse kinematics solver, we can achieve both the individualization and retargetting of characters in real time.

The Inverse Kinematics (IK) algorithms implemented in the open-source Orocos Kinematics and Dynamics Library (KDL) are arguably the most widely-used generic IK solvers worldwide. However, KDL's only joint-limit-constrained IK implementation, a pseudoinverse Jacobian IK solver, repeatedly exhibits false-negative failures on various humanoid platforms. In order to find a better IK solver for generic manipulator chains, a variety of open-source, drop-in alternatives have been implemented and evaluated for this paper. This article provides quantitative comparisons, using multiple humanoid platforms, between an improved implementation of the KDL inverse Jacobian algorithm, a set of sequential quadratic programming (SQP) IK algorithms that use a variety of quadratic error metrics, and a combined algorithm that concurrently runs the best performing SQP algorithm and the improved inverse Jacobian implementation. The best alternative IK implementation finds solutions much more often than KDL, is faster on average than KDL for typical manipulation chains, and (when desired) allows tolerances on each Cartesian dimension, further improving speed and convergence when an exact Cartesian pose is not possible and/or necessary.

We present a novel, hybrid neuro-genetic visuomotor architecture for object grasping on a humanoid robot. The approach combines the state-of-the-art object detector RetinaNet, a neural network-based coordinate transformation and a genetic-algorithm-based inverse kinematics solver. We claim that a hybrid neural architecture can utilise the advantages of neural and genetic approaches: while the neural components accurately locate objects in the robot’s three-dimensional reference frame, the genetic algorithm allows reliable motor control for the humanoid, despite its complex kinematics. The modular design enables independent training and evaluation of the components. We show that the additive error of the coordinate transformation and inverse kinematics solver is appropriate for a robotic grasping task. We additionally contribute a novel spatial-oversampling approach for training the neural coordinate transformation that overcomes the known issue of neural networks with extrapolation beyond training data and the extension of the genetic inverse kinematics solver with numerical fine-tuning. The grasping approach was realised and evaluated on the humanoid robot platform NICO in a simulation environment.

This paper presents a technique for adapting existing motion of a human-like character to have the desired features that are specified by a set of constraints. This problem can be typically formulated as a spacetime constraint problem. Our approach combines a hierarchical curve fitting technique with a new inverse kinematics solver. Using the kinematics solver, we can adjust the configuration of an articulated figure to meet the constraints in each frame. Through the fitting technique, the motion displacement of every joint at each constrained frame is interpolated and thus smoothly propagated to frames. We are able to adaptively add motion details to satisfy the constraints within a specified tolerance by adopting a multilevel Bspline representation which also provides a speedup for the interpolation. The performance of our system is further enhanced by the new inverse kinematics solver. We present a closed-form solution to compute the joint angles of a limb linkage. This analytical method greatly reduces the burden of a numerical optimization to find the solutions for full degrees of freedom of a human-like articulated figure. We demonstrate that the technique can be used for retargetting a motion to compensate for geometric variations caused by both characters and environments. Furthermore, we can also use this technique for directly manipulating a motion clip through a graphical interface. CR Categories: I.3.7 [Computer Graphics]: Threedimensional Graphics—Animation; G.1.2 [Numerical Analysis]: Approximation—Spline and piecewise polynomial approximation

It is undoubtful that artificial intelligence (AI) is being the trend of computer science and this trend is still ongoing in the far future even though technologies are being developed suddenly fast because computer science does not reach the limitation of approaching biological world yet. Machine learning (ML), which is a branch of AI, is a spearhead but not a key of AI because it sets first bricks to build up an infinitely long bridge from computer to human intelligence, but it is also vulnerable to environmental changes or input errors. There are three typical types of ML such as supervised learning, unsupervised learning, and reinforcement learning (RL) where RL, which is adapt progressively to environmental changes, can alleviate vulnerability of machine learning but only RL is not enough because the resilience of RL is based on iterative adjustment technique, not based on naturally inherent aspects like data mining approaches and moreover, mathematical fundamentals of RL lean forwards swing of stochastic process. Fortunately, artificial neural network, or neural network (NN) in short, can support all three types of ML including supervised learning, unsupervised learning, and RL where the implicitly regressive mechanism with high order through many layers under NN can improve the resilience of ML. Moreover, applications of NN are plentiful and multiform because three ML types are supported by NN; besides, NN training by backpropagation algorithm is simple and effective, especially for sample of data stream. Therefore, this study research is an introduction to NN with easily understandable explanations about mathematical aspects under NN as a beginning of stepping into deep learning which is based on multilayer NN. Deep learning, which is producing amazing results in the world of AI, is undoubtfully being both spearhead and key of ML with expectation that ML improved itself by deep learning will become both spearhead and key of AI, but this expectation is only for ML researchers because there are many AI subdomains are being invented and developed in such a way that we cannot understand exhaustedly. It is more important to recall that NN, which essentially simulates human neuron system, is appropriate to the philosophy of ML that constructs an infinitely long bridge from computer to human intelligence.

This paper discusses the problem of generating trajectory for a 2-dof parallel manipulator with actuator redundancy. We investigate how to transform the speed and torque limits of the actuators to the velocity and acceleration limits in the end-effector space where the desired trajectory is normally specified. How these limits vary over the workspace is also investigated in details for the particular platform. With these limits, the automatic trajectory generation algorithm for traditional x-y table is modified to suit our parallel manipulator with actuator redundancy. Our trajectory generation algorithm can automatically generate symmetric straight line point-to-point trapezoidal trajectory for the parallel manipulator given the start point, the end point, and the desired traveling time. If it is impossible for the parallel manipulator to travel the distance within the desired traveling time under the velocity and acceleration limits, the shortest time trajectory will be generated instead.

As humanoid robotics advances beyond bipedal walking, complex motions involving the whole body are necessary. Most recent humanoids represent the range of motion of the human spine with a single rotation joint. While this joint allows the body to swing during dynamic walking, any bending must be performed only with the legs. This paper develops a skewed rotation plane (SRP) waist joint to give a humanoid robot the same range of torso motion as a human. The SRP design reduces holding torque compared to a orthogonal-axis joint. An inverse kinematics solver using Jacobian Pseudo-inverse was developed to produce smooth torso orientation trajectories. Finally, a mechanical prototype developed and fitted to Drexel University's Jaemi Hubo to verify and validate the model.

In the framework of virtual prototyping, a virtual human is a useful tool to evaluate what one can reach or see. However because engineers and industrial designers are not necessarily skilled animators, it is important to provide intuitive or high-level interfaces to control virtual characters. With those considerations in mind, in this thesis we investigate three interfaces to control virtual humans for reaching tasks. First we propose an online full-body control using motion capture technology. The posture is reconstructed with an inverse kinematics solver. Coupled with a collision avoidance module, it is a powerful interface to control interactively virtual humans in cluttered environment. Secondly, we propose a novel interface for the full-body online control of differing-height characters. Instead of scaling the markers to the size of the virtual human, we suggest inversely scaling the virtual world. We conduct a user-study to evaluate if such interface is susceptible to help the user experiences the efforts involved by the differing-height character to complete a reaching task. Finally, we investigate motion graph technology for navigation coupled with reaching motions.

In recent years, hydraulic manipulators have been more and more widely used in the field of industrial automation because of their high power-to-weight ratio. Compared with autonomous control, bilateral teleoperation is a more reliable way to operate hydraulic manipulators on harsh occasions. However, teleoperation application in the hydraulic field is still limited due to operation accuracy and master–slave heterogeneous problem. In this paper, a teleoperation framework based on master–slave heterogeneous matching and nonlinear precise motion control for hydraulic manipulator is proposed, including workspace mapping algorithm, velocity-limited interpolator, inverse kinematics solver, trajectory planner, model-based controller, and feedback torque generator. The high-precision teleoperation of the hydraulic manipulator in the situation of master–slave heterogeneity is finally realized based on the designed teleoperation framework. The specialized experiment is set up, and the results also show that the method proposed in this paper can satisfy the teleoperation requirements of the hydraulic manipulator with the maximum operating accuracy up to 0.02 m under the condition of master–slave heterogeneity, which can greatly improve the applicability of hydraulic manipulator teleoperation in the case of master–slave heterogeneity.

Robotic assistance in nursing is becoming increasingly important due to labor shortage and novel capabilities in hard- and software. In nursing, patients often need to be manipulated which requires force distribution beyond the scope of traditional end effectors. This can be solved with whole arm manipulation (WAM). Unfortunately, commanding WAM configurations for grasping patients is very complex, as many contact points need to be achieved, and all 6 or 7 degrees of freedom have to be defined. In this paper, a novel interaction concept and according to inverse kinematic solvers based on a local elliptic approximation are proposed, which allows commanding a WAM holding grasp with one click on a point cloud. It is shown how the local elliptic model is used for formulating a very fast analytic multi-contact inverse kinematic solver. Two alternative solvers are derived by extending the BioIK method and by utilizing the popular sequential quadratic programming approach. Experiments show the efficiency of the analytic method and the robustness of the local elliptic model relative to perceptual perturbations.

Articulated hand tracking systems have been commonly used in virtual reality applications, including systems with human-computer interaction or interaction with game consoles. However, building an effective real-time hand pose tracker remains challenging. In this paper, we present a simple and efficient methodology for tracking and reconstructing 3d hand poses using a markered optical motion capture system. Markers were positioned at strategic points, and an inverse kinematics solver was incorporated to fit the rest of the joints to the hand model. The model is highly constrained with rotational and orientational constraints, allowing motion only within a feasible set. The method is real-time implementable and the results are promising, even with a low frame rate.

In order to grasp an object, we need to solve the inverse kinematics problem, i.e., the coordinate transformation from the visual coordinates to the joint angle vector coordinates of the arm. In human motion control, the learning of the hand position error feedback controller in the inverse kinematics solver is important. Although several models of coordinate transformation learning have been proposed, they suffer from a number of drawbacks. This paper proposes a novel model of the coordinate transformation learning of the human visual feedback controller by using disturbance noise and feedback error signal. The feasibility of the proposed model is illustrated using numerical simulations.

Many function-approaching methods such as neural network, fuzzy method are used in reinforcement learning methods for solving its huge problem space dimensions. This paper presents a novel ART2 neural network based reinforcement learning method (ART2-RL) to solve the space problem. Because of its adaptive resonance characteristic, ART2 neural network is used to process the space measurement of reinforcement learning and improve the learning speed. This paper also gives the reinforcement learning algorithm based on ART2. A simulation of path planning of mobile robot has been developed to prove the validity of ART2-RL. As the complexity of the simulation increased, the result shows that the number of collision between robot and obstacles is effectively decreased; the novel neural network model provides significant improvement in the space measurement of reinforcement learning.

Learning ability is the basic characteristic of human intelligence. The July 1, 2005 issue of Science published a list of 125 important questions in science. Among them, the question 94 “What are the limits of learning by machines?”. The annotation “Computers can already beat the world’s best chess players, and they have a wealth of information on the Web to draw on. But abstract reasoning is still beyond any machine”. In recent artificial intelligence has made great progresses. In 1997, the rise of the man-machine war, IBM Supercomputer Deep Blue defeated the chess master Garry Kasparov. On February 14, 2011, IBM’s Watson supercomputer won a practice round against Jeopardy champions Ken Jennings and Brad Rutter. In March 2016, Google DeepMind’s AlphaGo sealed a 4-1 victory over a South Korean Go grandmaster Lee Se-dol. This paper focuses on the machine learning methods of AlphaGo, including reinforcement learning, deep learning, deep reinforcement learning, analysis of the existing problems and the latest research progress. Deep reinforcement learning is the combination of deep learning and reinforcement learning, which can realize the learning algorithm from the perception to action. Simply said, this is the same as human behavior, input sensing information such as vision, and then, direct output action through the deep neural network. Deep reinforcement learning has the potential to learn a variety of skills for the robot to achieve full autonomy. Even though reinforcement learning is practiced successfully, but feature states need to manually set, for complex scene is a difficult thing, especially easy to cause the dimension disaster, and expression is not good. In 2010, Sascha Lange and Martin Riedmiller proposed deep auto-encoder neural networks in reinforcement learning to extract feature, which is used to control the visual correlation. In 2013, DeepMind proposed deep Q-network (DQN) in NIPS 2013, using convolution neural network to extract features, and then applied in reinforcement learning. They continue to improve and published an improved version of DQN on Nature in 2015, which has aroused widespread concern. In order to break through the limits of learning by machines, cognitive machine learning is proposed, which is the combination of machine learning and brain cognition, so that the machine intelligence is constantly evolving, and gradually reaches the human level of artificial intelligence. A cognitive model entitled Consciousness And Memory (CAM) is proposed by author, which consists of memory, consciousness, high-level cognitive functions, perception and motor. High-level cognitive functions of the brain include learning, language, thinking, decision making, emotion, and so on. Learning is a course to accept the stimulus through the nervous system and obtain new behavior, habits and accumulation experience. According to the current research progress of brain science and cognitive science, cognitive machine learning may be interested in learning emergence, procedural memory knowledge learning, learning evolution and so on. For intelligence, so-called evolution is refers to the learning of learning and the structure also follows the change. It is important to record the learning result by structure changing and improve the learning method.