Study on Application of Machine Learning Intelligence for Identification and Control of Nonlinear Dynamical Systems: Case Study Robotic Manipulators

In the literature, machine learning has been referred to as deeply structured learning, hierarchical learning and feature based learning which can provide computational models from high-level data abstractions. One of the most used learning structures is the multiple-layered models of inputs, commonly known as neural networks, which comprise multiple levels of non-linear operations. The machine learning algorithms are able solve many problems around fault detection, isolation and recovery. There has been a growing interest in using learning architectures in advanced robotics applications, e.g., object detection, scene semantic segmentation, and grasping. The real-time learning of high-dimensional features for robotics applications via machine learning techniques is another important topic. In addition, other topics in robotics such as obstacle detection and context-dependent social mapping are also being addressed by researchers through machine learning methods. Machine learning algorithms provide real time driving decisions for automated vehicles (self-driving vehicles or driverless cars) from integration of numerous sensors onboard the vehicle. The advancement of autonomous navigation and situational awareness systems adapt neural networks for analyzing the multi-modal sensor inputs. We observe that machine learning algorithms influence largely in decision making process. But, there is need to understand the control system consequences for adapting the outcome of the machine learning algorithm. This proposal presents the detailed study on the influences of machine learning architectures and algorithms for modeling and control of nonlinear dynamical system. Research Outcome: · Knowledge on machine learning architectures (Support Vector Machines (SVMs), Conditional Random Field, supervised neural network) · Understanding the constraints on applicability of ML architectures for nonlinear dynamical system · Study on real time control of nonlinear dynamical system with ML algorithm in closed loop.