Encyclopedia
Please note this is an old version of this entry, which may differ significantly from the current revision.
Subjects:
Robotics
Robot arm motion control is a fundamental aspect of robot capabilities, with arm reaching ability serving as the foundation for complex arm manipulation tasks. However, traditional inverse kinematics-based methods for robot arm reaching struggle to cope with the increasing complexity and diversity of robot environments, as they heavily rely on the accuracy of physical models.
- arm reaching
- motion planning
- inner rehearsal
- internal model
- human cognitive mechanism
1. Introduction
In recent years, robots have played important roles in many fields, especially for humanoid robots. As arm manipulation is one of the most basic abilities for human beings [1], arm motion control is also an indispensable ability for humanoid robots [2]. In modern factories, automation manufacturing and many other production activities are inseparable from robot arms [3]. Among various types of arm manipulation abilities, arm reaching is one of the most basic, and it is the first step in many complex arm motions, such as grasping and placing, and can also lay the foundation for subsequent motion, perception, and cognition [4,5].
The major goal of robot arm reaching is to choose a set of appropriate arm joint angles so that the end-effectors can reach the target position in Cartesian space with a certain posture, which is commonly referred to as internal model control (IMC) [6]. In 1998, Wolpert et al. [7] reviewed the necessity of such an internal model and the evidence in a neural loop. In the same year, Wolpert and Kawato [8] proposed a new architecture based on multiple pairs of inverse (controller) and forward (predictor) models, where the inverse and forward models are tightly coupled. researchers usually implement robots’ motion control through kinematic or dynamic modeling [9]. However, since the control of force and torque is not involved in robot arm reaching tasks, only the kinematic method is considered. The internal kinematics model can be separated into the inverse kinematics (IK) model and forward kinematics (FK) model according to the input and output of the model.
In the task of reaching, the major problem is how to build an accurate IK model that maps a certain posture to a set of joint angles. There are mainly two types of approaches to robot arm reaching control: conventional IK-based approaches and learning-based approaches. Table 1 shows some related research.
Table 1. Classification of robot arm reaching control approaches.
| Classification | Approaches |
|---|---|
| Numerical method [10] | |
| Conventional IK-based | Analytical method [11,12] |
| Geometric method [13] | |
| Learning-based | Supervised learning: deep neural networks [14], spiking neural networks [15] Unsupervised learning: self-organizing maps [16], reinforcement learning [17,18] |
In conventional IK-based approaches, the mapping from the pose of the end-effector to the joint angles is built based on the mechanical structure of the robot’s arm. The joint angles are calculated with an analytical method or iterative method [19]. In this type of approach, the accuracy of the IK model strongly depends on the measurement accuracy of the robot’s mechanical parameters, and researchers need to solve equations in high dimensions [20]. This brings difficulties in calibrating the parameters, which may change continuously because of the wear and tear of the robot. The application of conventional IK-based approaches in complex and unstructured environments is limited by the accuracy of the measurement and the manual calibration of the parameters.
To avoid the drawbacks of conventional IK-based approaches, researchers have focused on learning methodologies to manipulate robot arms, rather than control-theoretic approaches, in recent years.
It is recognized that robots must have the ability to learn continuously in order to adapt to the ever-changing and unstructured environment. Learning can also decrease the complexity of control for robots with large degrees of freedom [21]. The learning-based approaches are inspired by cognitive, motion, and other relevant mechanisms in human beings. The inverse model is established by means of self-exploration, based on neural networks [22], reinforcement learning [18], or other learning algorithms.
Using a learning-based model, accurate measurements of the robot’s mechanical parameters are no longer the decisive factor in arm reaching. However, a well-performing model requires a large amount of training data, mainly generated from trial-and-error experiments, which might cause great abrasion to the robot. Many researchers train the model in simulation first and then refine the model in a real robot platform [23].
To perform well in robot arm reaching tasks, it is necessary to ensure the accuracy of the internal model and the target positioning accuracy as well. As discussed above, in arm reaching manipulation, the joint angles can be calculated by the inverse kinematics model in IK-based approaches or the inverse model built through learning-based approaches once researchers know the position and posture of the end-effector, while the target pose is mainly determined through visual positioning, which is strongly related to the performance of the camera.
Compared with the absolute positioning method, relative positioning will help to reduce the influence of perception error. Research also shows that in the process of human arm reaching, older children and adults consider both target and hand positions [24,25]. Based on this mechanism, Luo et al. [26] proposed a relative-location-based approximation approach in 2018. In their work, the reaching process is divided into two stages, rough reaching and iterative adjustment. However, the motion commands are combined with six basic moving directions, which may lead to non-smooth reaching trajectories. To smooth the trajectory of the reaching trajectory, researchers use differential theory for reference and limit the distance of each movement to a small-scale given threshold.
The reaching approaches described in the above research are mainly in an open-loop manner [27,28]. Each joint of the robot arm moves to a target angle calculated by the inverse model, where feedbacks are less considered. However, when it comes to the real environment, the planning becomes more difficult and less robust [29], and the motion commands generated by the inverse model may not be executed precisely because of mechanical errors. Thus, feedbacks on the execution and corresponding motion planning are needed to increase the reaching performance. In 2017, Krivic et al. [30] proposed a planning approach to predict new information about a partially known state and to decrease uncertainty. In 2019, Luo et al. [31] proposed an action selection method based on sensorimotor prediction and anticipating the consequences of currently executable actions in internal simulation.
2. Reaching with Visual Servoing
The general aim of the visual servo system in reaching tasks is to reduce the error 𝑒(𝑡) [33], defined as
where 𝑚(𝑡) represents image measurements, a indicates any potential additional data, and 𝑠∗ stores the desired final position.
As an effective method, visual servoing has been applied in motion estimation [34], position control [35], and other robotics tasks. In 2000, Shen et al. proved that with visual servoing, the robot trajectory can approach the desired trajectory asymptotically [35]. In 2013, Lampe et al. proposed a reinforcement learning system for autonomous robot arm reaching using visual servoing. They used a visual feedback control loop to realize control, making it both reactive and robust to noise [17].
3. Learning-Based Internal Model
In 1987, Kosuge et al. introduced the concept of a virtual internal model and applied it in robot arm control [36]. In 1993, Koivisto et al. analyzed a nonlinear internal model controller realized with a multilayer perceptron neural network and proved that implementing practical experience with neural networks can provide IMC robust performance [37], demonstrating that a learning-based internal model is reliable.
In recent years, the learning-based internal model has been used in ethical robots [38], robot intent prediction [39], robot manipulation [18], and other robotics research; it simulates the cognitive mechanisms of humans and makes robots more intelligent.
4. Inner Rehearsal
Humans can simulate the normal execution results of an action through inner rehearsal [40]. Taking advantage of the inner rehearsal mechanism, people can try to run the potential actions in their minds and predict their results, such as actions or decisions that are not explicitly executed [41,42].
For robots, through inner rehearsal, the result of a motion command can be predicted without actual execution [32]. Inner rehearsal has been used in robot linguistic interpretation [43], relation learning [44], and navigation [45] to predict the result of a command and choose an action accordingly. In this way, robots can avoid unnecessary attempts and conduct the best move. In their recent work, Atkinson et al. proposed to use pseudo-rehearsal in reinforcement learning to prevent neural networks from catastrophic forgetting [46,47].
This entry is adapted from the peer-reviewed paper 10.3390/biomimetics8060491
This entry is offline, you can click here to edit this entry!
