China
Africa
Explore chapters and articles related to this topic
Machine Learning
Published in Seyedeh Leili Mirtaheri, Reza Shahbazian, Machine Learning Theory to Applications, 2022
Seyedeh Leili Mirtaheri, Reza Shahbazian
Reinforcement learning solves the hard problem of correlating immediate actions with the delayed outcomes they produce. Like humans, reinforcement learning algorithms sometimes have to wait to see the result of their decisions. They operate in a delayed return environment, where it can be difficult to understand which action leads to which outcome over many time steps. Like the baby learns how to walk during the time, the reinforcement learning algorithms are slowly performing better and better in more ambiguous, real world environments while choosing from an arbitrary number of possible actions, rather than from the limited options of a repeatable video game. That is, they are beginning to achieve goals in the real world. DeepMind claimed in May 2021 that reinforcement learning was probably sufficient to achieve artificial general intelligence. Also, companies are starting to apply deep reinforcement learning to problems in industry especially in the robotics industry. For example, Pieter Abbeel’s Covariant uses deep reinforcement learning in industrial robotics. Pathmind applies deep reinforcement learning to simulations of industrial operations and supply chains to optimize factories, warehouses and logistics. Google is applying deep reinforcement learning to problems such as robot locomotion and chip design, while Microsoft relies on deep reinforcement learning to power its autonomous control systems technology.
Robotic Swarms for Mine Detection System of Systems Approach
Published in Thrishantha Nanayakkara, Ferat Sahin, Mo Jamshidi, Intelligent Control Systems with an Introduction to System of Systems Engineering, 2018
Thrishantha Nanayakkara, Ferat Sahin, Mo Jamshidi
Our MSR approach constitutes software modularity that matches the hardware modularity mentioned above. In our approach, each hardware module corresponds to a software module as well, instead of each hardware layer dealing with only the hardware components, as in OSCAR. An object-oriented paradigm was used to design our modular software methodologies. The approach has abstract classes and their concrete subclasses. The abstract classes are robot, locomotion, control, communication, sensor, and actuators. The concrete classes are derived from these abstract classes. For example, a wireless layer may be instantiated from the communication class. The wireless object represents the robot’s communication module. The concrete classes inherit functionality from the abstract classes, in addition to their hardware-specific functionalities. Even though the modules contain their own software modules, the central controller (or OS) identifies newly inserted modules and creates the corresponding objects and interfaces to the hardware module. Figure 12.2 shows the class structure for the modular software architecture.
Robotics
Published in Jian Chen, Bingxi Jia, Kaixiang Zhang, Multi-View Geometry Based Visual Perception and Control of Robotic Systems, 2018
Jian Chen, Bingxi Jia, Kaixiang Zhang
A rigid body in physical space can be described by the position and orientation, which are collectively named as the pose information. Then, the robot locomotion can be described by the pose information, which serves as the output of the robot model. The motion of a rigid body consists of translations and rotations, resulting in the position and orientation descriptions with respect to the reference coordinate system. As shown in Figure 1.1, two coordinate frames F and F′ exist in 3D Euclidean space. The motion from F to F′ can be described by a translation and a rotation. Then, the pose of frame F′ with respect to frame F can be described by the position and the orientation.
Path tracking control of a snake robot with a passive joint
Published in Advanced Robotics, 2023
Kazunori Sakakibara, Ryo Ariizumi, Toru Asai, Shun-ichi Azuma
Among many problems on the snake robot, path-following control on a plane is particularly important because it allows the snake robot to move along an arbitrary path. Path-following control is, therefore, relevant to the future role of snake robots, such as patrolling a certain path or mapping the road environment. Path-following control of a snake robot on the plane has been studied by several authors. It has been shown that a snake robot can reproduce the general movement patterns of biological snakes by following a shape called a serpenoid curve in its posture [8]. It is also known that the direction of robot locomotion can be controlled by adding an offset to each joint angle in that case [9]. Using this directional control, path-following control is achieved by determining the robot's direction of travel according to the line-of-sight (LOS) guidance law [10]. Others have studied path-following control described by a simplified model that uses cascade system theory to convert joint angles into lateral displacements [11].
Emergent control in the context of industry 4.0
Published in International Journal of Computer Integrated Manufacturing, 2022
In Fuentes et al. (2015), the hexapod robot locomotion is studied using central pattern generators (GPC), generating simple signals to coordinate movements at the micro-level, which are models with biologically inspired algorithms, such as forward-fed neural networks. CPGs can also be considered as coupled nonlinear oscillator systems, that is, they are biological neural circuits that produce rhythmic outputs in the absence of rhythmic inputs. Examples are found in motor behaviors, such as walking, swimming, flying or breathing of organisms alive. They demonstrate how chaotic controlled oscillators, when used with local sensory feedback, promote the emergence of adaptive robot locomotion under conditions of terrain uncertainty. This type of CPG favors distributed control approaches through simple control signals. Its simple structure also facilitates the integration of sensory information when the CPGs constituted of coupled nonlinear oscillators are applied to the control of robotic locomotion.
Effects of passive and active joint compliance in quadrupedal locomotion
Published in Advanced Robotics, 2018
M. Mutlu, S. Hauser, A. Bernardino, A. J. Ijspeert
The driving force this paper is the increasing need to understand role of both passive and active compliance better in quadrupedal locomotion. To this end, a compliant modular quadruped robot has been designed using low-budget off-the-shelf components. The robot is highly customizable and fast to reconfigure which enables a wide set of experiment possibilities involving morphological changes. The goal of the paper is to answer a set of questions which are How does the passive compliance of legs affect quadrupedal robot locomotion?Can having asymmetric passive compliance on fore and hind limbs increase the performance of locomotion?Is it possible to boost adaptation of the robot to its environment using active compliance?Does active compliance and passive compliance cooperate well or destruct each other's contributions?Do results scale up to different terrains?