Learning how to grasp based on neural network retraining

Humans have an incredible capacity to manipulate objects using dextrous hands. A large number of studies indicate that robot learning by demonstration is a promising strategy to improve robotic manipulation and grasping performance. Concerning this subject we can ask: How does a robot learn how to grasp? This work presents a method that allows a robot to learn new grasps. The method is based on neural network retraining. With this approach we aim to enable a robot to learn new grasps through a supervisor. The proposed method can be applied for 2D and 3D cases. Extensive object databases were generated to evaluate the method performance in both 2D and 3D cases. A total of 8100 abstract shapes were generated for 2D cases and 11700 abstract shapes for 3D cases. Simulation results with a computational supervisor show that a robotic system can learn new grasps and improve its performance through the proposed HRH (Hopfield-RBF-Hopfield) grasp learning approach.     

Motion Planning for Two Robots of an Object Handling System Considering Fast Transition Between Stable States

Fast transition from a stable initial state to a stable handling state is important when multiple mobile robots grasp and transport a bulky and heavy object. In this paper, we present motion planning for two robots of an irregularly shaped object handling system considering fast transition between stable states. A cooperative object handling system consisting of a gripper robot equipped with a gripper and a lifter robot equipped with a lifter was first designed. Then, a strategy to realize fast transition between stable states by using the object handling system designed was proposed. While grasping and lifting an object off the ground, a gripper robot grasps and lifts up the object from one side to provide enough space for a lifter robot to lift the object off the ground cooperatively. Fast transition between stable states is formulated as a constraint optimization problem. The goal is to realize transition from a stable initial state to a stable handling state in a minimal amount of time. Experiments involving two robots and everyday objects were conducted. The two robots cooperatively obtained fast transition between stable states. The results illustrate the validity of the proposed method.     

Towards cognitive grasping: modeling of unknown objects and its corresponding grasp types

This paper describes an intuitive approach for a cognitive grasp of a robot. The cognitive grasp means the chain of processes that make a robot to learn and execute a grasping method for unknown objects like a human. In the learning step, a robot looks around a target object to estimate the 3D shape and understands the grasp type for the object through a human demonstration. In the execution step, the robot correlates an unknown object to one of known grasp types by comparing the shape similarity of the target object based on previously learned models. For this cognitive grasp, we mainly deal with two functionalities such as reconstructing an unknown 3D object and classifying the object by grasp types. In the experiment, we evaluate the performance of object classification according to the grasp types for 20 objects via human demonstration.     

Distributed cooperative load redistribution for fault tolerance in a team of four object-lifting robots

Fault tolerance in a team of cooperative and distributed object-lifting robots is dealt with. It is assumed that one of four object lifting robots misses a portion of its lifting power and the robots must redistribute the load among themselves to perform their task. Two distributed and cooperative methods for load reallocation among the position-controlled robots without requiring them to change their grasp positions are introduced. The first method benefits from the existing redundancy in the number and lifting power of robots. In the second method, the object is tilted in order to move the zero moment point (ZMP) away from the faulty robot and, consequently, redistributing the load. Difficulties in controlling ZMP movements are pointed out. Therefore, the second fault-clearing procedure is designed such that the ZMP position is controlled without resorting to sophisticated or centralized control algorithms. Stability of the proposed methods is mathematically proven and the deadlocks are investigated. It is also noted that the required sensory system and robot behavior in the proposed strategies are exactly the same as those used in the object-lifting task. Consequently, no additional complexity is imposed on the system. The basic idea in ALLIANCE is used for developing a mechanism for each robot to process help requests and select proper actions in a distributed fashion without negotiating with its teammates. Simulation results are given to support the developed methods.     

Dyna-Q-based vector direction for path planning problem of autonomous mobile robots in unknown environments

Reinforcement learning (RL) is a popular method for solving the path planning problem of autonomous mobile robots in unknown environments. However, the primary difficulty faced by learning robots using the RL method is that they learn too slowly in obstacle-dense environments. To more efficiently solve the path planning problem of autonomous mobile robots in such environments, this paper presents a novel approach in which the robot’s learning process is divided into two phases. The first one is to accelerate the learning process for obtaining an optimal policy by developing the well-known Dyna-Q algorithm that trains the robot in learning actions for avoiding obstacles when following the vector direction. In this phase, the robot’s position is represented as a uniform grid. At each time step, the robot performs an action to move to one of its eight adjacent cells, so the path obtained from the optimal policy may be longer than the true shortest path. The second one is to train the robot in learning a collision-free smooth path for decreasing the number of the heading changes of the robot. The simulation results show that the proposed approach is efficient for the path planning problem of autonomous mobile robots in unknown environments with dense obstacles.     

Robotic grasping and manipulation through human visuomotor learning

A major goal of robotics research is to develop techniques that allow non-experts to teach robots dexterous skills. In this paper, we report our progress on the development of a framework which exploits human sensorimotor learning capability to address this aim. The idea is to place the human operator in the robot control loop where he/she can intuitively control the robot, and by practice, learn to perform the target task with the robot. Subsequently, by analyzing the robot control obtained by the human, it is possible to design a controller that allows the robot to autonomously perform the task. First, we introduce this framework with the ball-swapping task where a robot hand has to swap the position of the balls without dropping them, and present new analyses investigating the intrinsic dimension of the ball-swapping skill obtained through this framework. Then, we present new experiments toward obtaining an autonomous grasp controller on an anthropomorphic robot. In the experiments, the operator directly controls the (simulated) robot using visual feedback to achieve robust grasping with the robot. The data collected is then analyzed for inferring the grasping strategy discovered by the human operator. Finally, a method to generalize grasping actions using the collected data is presented, which allows the robot to autonomously generate grasping actions for different orientations of the target object.     

Inverse reinforcement learning-based time-dependent A* planner for human-aware robot navigation with local vision

In an environment where robots coexist with humans, mobile robots should be human-aware and comply with humans' behavioural norms so as to not disturb humans' personal space and activities. In this work, we propose an inverse reinforcement learning-based time-dependent A* planner for human-aware robot navigation with local vision. In this method, the planning process of time-dependent A* is regarded as a Markov decision process and the cost function of the time-dependent A* is learned using the inverse reinforcement learning via capturing humans' demonstration trajectories. With this method, a robot can plan a path that complies with humans' behaviour patterns and the robot's kinematics. When constructing feature vectors of the cost function, considering the local vision characteristics, we propose a visual coverage feature for enabling robots to learn from how humans move in a limited visual field. The effectiveness of the proposed method has been validated by experiments in real-world scenarios: using this approach robots can effectively mimic human motion patterns when avoiding pedestrians; furthermore, in a limited visual field, robots can learn to choose a path that enables them to have the larger visual coverage which shows a better navigation performance.     

基于人工神经网络的多机器人协作学习研究

机器人足球比赛是一个有趣并且复杂的新兴的人工智能研究领域，它是一个典型的多智能体系统。文中主要研究机器人足球比赛中的协作行为的学习问题，采用人工神经网络算法实现了两个足球机器人的传球学习，实验结果表明了该方法的有效性。最后讨论了对BP算法的诸多改进方法。     

Adaptation technique for integrating genetic programming and reinforcement learning for real robots

We propose an integrated technique of genetic programming (GP) and reinforcement learning (RL) to enable a real robot to adapt its actions to a real environment. Our technique does not require a precise simulator because learning is achieved through the real robot. In addition, our technique makes it possible for real robots to learn effective actions. Based on this proposed technique, we acquire common programs, using GP, which are applicable to various types of robots. Through this acquired program, we execute RL in a real robot. With our method, the robot can adapt to its own operational characteristics and learn effective actions. In this paper, we show experimental results from two different robots: a four-legged robot "AIBO" and a humanoid robot "HOAP-1." We present results showing that both effectively solved the box-moving task; the end result demonstrates that our proposed technique performs better than the traditional Q-learning method.     

Adaptive neural controller for cooperative multiple robot manipulator system manipulating a single rigid object

In this article, an adaptive neural controller is developed for cooperative multiple robot manipulator system carrying and manipulating a common rigid object. In coordinated manipulation of a single object using multiple robot manipulators simultaneous control of the object motion and the internal force exerted by manipulators on the object is required. Firstly, an integrated dynamic model of the manipulators and the object is derived in terms of object position and orientation as the states of the derived model. Based on this model, a controller is proposed that achieves required trajectory tracking of the object as well as tracking of the desired internal forces arising in the system. A feedforward neural network is employed to learn the unknown dynamics of robot manipulators and the object. It is shown that the neural network can cope with the unknown nonlinearities through the adaptive learning process and requires no preliminary offline learning. The adaptive learning algorithm is derived from Lyapunov stability analysis so that both error convergence and tracking stability are guaranteed in the closed loop system. Finally, simulation studies and analysis are carried out for two three-link planar manipulators moving a circular disc on specified trajectory.     

Telemanipulation of cooperative robots: a case of study

This article addresses the problem of dexterous robotic grasping by means of a telemanipulation system composed of a single master and two slave robot manipulators. The slave robots are analysed as a cooperative system where it is assumed that the robots can push but not pull the object. In order to achieve a stable rigid grasp, a centralised adaptive position-force control algorithm for the slave robots is proposed. On the other hand, a linear velocity observer for the master robot is developed to avoid numerical differentiation. A set of experiments with different human operators were carried out to show the good performance and capabilities of the proposed control-observer algorithm. In addition, the dynamic model and closed-loop dynamics of the telemanipulation is presented.     

动态环境中的多机器人协同搬运

在多机器人协同搬运过程中,针对传统的强化学习算法仅使用数值分析却忽略了推理环节的问题,将多机器人的独立强化学习与“信念-愿望-意向”（BDI）模型相结合,使得多机器人系统拥有了逻辑推理能力,并且,采用距离最近原则将离障碍物最近的机器人作为主机器人,并指挥从机器人运动,提出随多机器人系统位置及最近障碍物位置变化的评价函数,同时将其与基于强化学习的行为权重结合运用,在多机器人通过与环境不断交互中,使行为权重逐渐趋向最佳。仿真实验表明,该方法可行,能够成功实现协同搬运过程。     

Acquiring Mobile Robot Behaviors by Learning Trajectory Velocities

The development of robots that learn from experience is a relentless challenge confronting artificial intelligence today. This paper describes a robot learning method which enables a mobile robot to simultaneously acquire the ability to avoid objects, follow walls, seek goals and control its velocity as a result of interacting with the environment without human assistance. The robot acquires these behaviors by learning how fast it should move along predefined trajectories with respect to the current state of the input vector. This enables the robot to perform object avoidance, wall following and goal seeking behaviors by choosing to follow fast trajectories near: the forward direction, the closest object or the goal location respectively. Learning trajectory velocities can be done relatively quickly because the required knowledge can be obtained from the robot's interactions with the environment without incurring the credit assignment problem. We provide experimental results to verify our robot learning method by using a mobile robot to simultaneously acquire all three behaviors.     

Cooperative strategy based on adaptive Q-learning for robot soccer systems

The objective of this paper is to develop a self-learning cooperative strategy for robot soccer systems. The strategy enables robots to cooperate and coordinate with each other to achieve the objectives of offense and defense. Through the mechanism of learning, the robots can learn from experiences in either successes or failures, and utilize these experiences to improve the performance gradually. The cooperative strategy is built using a hierarchical architecture. The first layer of the structure is responsible for assigning each role, that is, how many defenders and sidekicks should be played according to the positional states. The second layer is for the role assignment related to the decision from the previous layer. We develop two algorithms for assignment of the roles, the attacker, the defenders, and the sidekicks. The last layer is the behavior layer in which robots execute their behavior commands and tasks based on their roles. The attacker is responsible for chasing the ball and attacking. The sidekicks are responsible for finding good positions, and the defenders are responsible for defending competitor scoring. The robots' roles are not fixed. They can dynamically exchange their roles with each other. In the aspect of learning, we develop an adaptive Q-learning method which is modified form the traditional Q-learning. A simple ant experiment shows that Q-learning is more effective than the traditional techniques, and it is also successfully applied to the learning of the cooperative strategy.     

Combining active learning and reactive control for robot grasping

Grasping an object is a task that inherently needs to be treated in a hybrid fashion. The system must decide both where and how to grasp the object. While selecting where to grasp requires learning about the object as a whole, the execution only needs to reactively adapt to the context close to the grasp’s location. We propose a hierarchical controller that reflects the structure of these two sub-problems, and attempts to learn solutions that work for both. A hybrid architecture is employed by the controller to make use of various machine learning methods that can cope with the large amount of uncertainty inherent to the task. The controller’s upper level selects where to grasp the object using a reinforcement learner, while the lower level comprises an imitation learner and a vision-based reactive controller to determine appropriate grasping motions. The resulting system is able to quickly learn good grasps of a novel object in an unstructured environment, by executing smooth reaching motions and preshaping the hand depending on the object’s geometry. The system was evaluated both in simulation and on a real robot.     

Autonomous reconfiguration of robot shape by using Q-learning

A modular robot can be built with a shape and function that matches the working environment. We developed a four-arm modular robot system which can be configured in a planar structure. A learning mechanism is incorporated in each module constituting the robot. We aim to control the overall shape of the robot by an accumulation of the autonomous actions resulting from the individual learning functions. Considering that the overall shape of a modular robot depends on the learning conditions in each module, this control method can be treated as a dispersion control learning method. The learning object is cooperative motion between adjacent modules. The learning process proceeds based on Q-learning by trial and error. We confirmed the effectiveness of the proposed technique by computer simulation.     

SSPQL: Stochastic shortest path-based Q-learning

Reinforcement learning (RL) has been widely used as a mechanism for autonomous robots to learn state-action pairs by interacting with their environment. However, most RL methods usually suffer from slow convergence when deriving an optimum policy in practical applications. To solve this problem, a stochastic shortest path-based Q-learning (SSPQL) is proposed, combining a stochastic shortest path-finding method with Q-learning, a well-known model-free RL method. The rationale is, if a robot has an internal state-transition model which is incrementally learnt, then the robot can infer the local optimum policy by using a stochastic shortest path-finding method. By increasing state-action pair values comprising of these local optimum policies, a robot can then reach a goal quickly and as a result, this process can enhance convergence speed. To demonstrate the validity of this proposed learning approach, several experimental results are presented in this paper.     

A morphable template framework for robot learning by demonstration: Integrating one-shot and incremental learning approaches

Robot learning by demonstration is key to bringing robots into daily social environments to interact with and learn from human and other agents. However, teaching a robot to acquire new knowledge is a tedious and repetitive process and often restrictive to a specific setup of the environment. We propose a template-based learning framework for robot learning by demonstration to address both generalisation and adaptability. This novel framework is based upon a one-shot learning model integrated with spectral clustering and an online learning model to learn and adapt actions in similar scenarios. A set of statistical experiments is used to benchmark the framework components and shows that this approach requires no extensive training for generalisation and can adapt to environmental changes flexibly. Two real-world applications of an iCub humanoid robot playing the tic-tac-toe game and soldering a circuit board are used to demonstrate the relative merits of the framework.     

Improving human robot collaboration through Force/Torque based learning for object manipulation

Human–Robot Collaboration (HRC) is a term used to describe tasks in which robots and humans work together to achieve a goal. Unlike traditional industrial robots, collaborative robots need to be adaptive; able to alter their approach to better suit the situation and the needs of the human partner. As traditional programming techniques can struggle with the complexity required, an emerging approach is to learn a skill by observing human demonstration and imitating the motions; commonly known as Learning from Demonstration (LfD). In this work, we present a LfD methodology that combines an ensemble machine learning algorithm (i.e. Random Forest (RF)) with stochastic regression, using haptic information captured from human demonstration. The capabilities of the proposed method are evaluated using two collaborative tasks; co-manipulation of an object (where the human provides the guidance but the robot handles the objects weight) and collaborative assembly of simple interlocking parts. The proposed method is shown to be capable of imitation learning; interpreting human actions and producing equivalent robot motion across a diverse range of initial and final conditions. After verifying that ensemble machine learning can be utilised for real robotics problems, we propose a further extension utilising Weighted Random Forest (WRF) that attaches weights to each tree based on its performance. It is then shown that the WRF approach outperforms RF in HRC tasks.     

Acquisition of joint attention through natural interaction utilizing motion cues

Joint attention is one of the most important cognitive functions for the emergence of communication not only between humans, but also between humans and robots. In previous work, we have demonstrated how a robot can acquire primary joint attention behavior (gaze following) without external evaluation. However, this method needs the human to tell the robot when to shift its gaze. This paper presents a method that does not need such a constraint by introducing an attention selector based on a measure consisting of saliencies of object features and motion cues. In order to realize natural interaction, a self-organizing map for real-time face pattern separation and contingency learning for gaze following without external evaluation are utilized. The attention selector controls the robot gaze to switch often from the human face to an object and vice versa, and pairs of a face pattern and a gaze motor command are input to the contingency learning. The motion cues are expected to reduce the number of incorrect training data pairs due to the asynchronous interaction that affects the convergence of the contingency learning. The experimental result shows that gaze shift utilizing motion cues enables a robot to synchronize its own motion with human motion and to learn joint attention efficiently in about 20 min.