Trajectory forecasting is used in many fields such as vehicle future trajectory prediction, stock market price prediction, human motion prediction and so on. Also, robots having the capability to reason about human behavior is an important aspect in human robot interaction. In trajectory prediction with regards to human motion prediction, implicit learning and reproduction of human behavior is the major challenge. This work tries to compare some of the recent advances taking a phenomenological approach to trajectory prediction. \par The work is expected to mainly target on generating future events or trajectories based on the previous data observed across many time intervals. In particular, this work presents and compares machine learning models to generate various human handwriting trajectories. Although the behavior of every individual is unique, it is still possible to broadly generalize and learn the underlying human behavior from the current observations to predict future human writing trajectories. This enables the machine or the robot to generate future handwriting trajectories given an initial trajectory from the individual thus helping the person to fill up the rest of the letter or curve. This work tests and compares the performance of Conditional Variational Autoencoders and Sinusoidal Representation Network models on handwriting trajectory prediction and reconstruction.

With the extensive technological progress made in the areas of drives, sensors and processing, exoskeletons and other wearable devices have become more feasible. However, the stringent requirements in regards to size and weight continue to exert a strong influence on the system-wide design of these devices and present many obstacles to a successful solution. On the other hand, while the area of controls has seen a significant amount of progress, there also remains a large potential for improvements. This dissertation approaches the design and control of wearable devices from a systems perspective and provides a framework to successfully overcome the often-encountered obstacles with optimal solutions. The electronics, drive and control system design for the HeSA hip exoskeleton project and APEx hip exoskeleton project are presented as examples of how this framework is used to design wearable devices. In the area of control algorithms, a real-time implementation of the Fast Fourier Transform (FFT) is presented as an alternative approach to extracting amplitude and frequency information of a time varying signal. In comparison to the peak search method (PSM), the FFT allows extracting basic gait signal information at a faster rate because time windows can be chosen to be less than the fundamental gait frequency. The FFT is implemented on a 16-bit processor and the results show the real-time detection of amplitude and frequency coefficients at an update rate of 50Hz. Finally, a novel neural networks based approach to detecting human gait activities is presented. Existing neural networks often require vast amounts of data along with significant computer resources. Using Neural Ordinary Differential Equations (Neural ODEs) it is possible to distinguish between seven different daily activities using a significantly smaller data set, lower system resources and a time window of only 0.1 seconds.

As the world moves towards faster production times, quicker shipping, and overall, more demanding schedules, the humans caught in the loop are subject to physical duress causing them to physically break down and have muscular skeletal injuries. Surprisingly, with more automation in logistics houses, the remaining workers must be quicker and do more, again resulting in muscular-skeletal injuries. To help alleviate this strain, a class of robotics and wearables has arisen wherein the human is assisted by a worn mechanical device. These devices, traditionally called exoskeletons, fall into two general categories: passive and active. Passive exoskeletons employ no electronics to activate their assistance and instead typically rely on the spring-like qualities of many materials. These are generally lighter weight than their active counterparts, but also lack the assistive power and can even interfere in other routine operations. Active exoskeletons, on the other hand, aim to avoid as much interference as possible by using electronics and power to assist the wearer. Properly executed, this can deliver power at the most opportune time and disengage from interference when not needed. However, if the tuning is mismatched from the human, it can unintentionally increase loads and possibly lead to other future injuries or harm. This dissertation investigates exoskeleton technology from two vantage points: the designer and the consumer. In the first, the creation of the Aerial Porter Exoskeleton (APEx) for the US Air Force (USAF). Testing of this first of its kind exoskeleton revealed a peak metabolic savings of 8.13% as it delivers 30 N-m of torque about each hip. It was tested extensively in live field conditions over 8 weeks to great success. The second section is an exploration of different commercially available exoskeletons and the development of a common set of standards/testing protocols is described. The results show a starting point for a set of standards to be used in a rapidly growing sector.

Reinforcement Learning(RL) algorithms have made a remarkable contribution in the eld of robotics and training human-like agents. On the other hand, Evolutionary Algorithms(EA) are not well explored and promoted to use in the robotics field. However, they have an excellent potential to perform well. In thesis work, various RL learning algorithms like Q-learning, Deep Deterministic Policy Gradient(DDPG), and Evolutionary Algorithms(EA) like Harmony Search Algorithm(HSA) are tested for a customized Penalty Kick Robot environment. The experiments are done with both discrete and continuous action space for a penalty kick agent. The main goal is to identify which algorithm suites best in which scenario. Furthermore, a goalkeeper agent is also introduced to block the ball from reaching the goal post using the multiagent learning algorithm.

The inherent behavior of many real world applications tends to exhibit complex or chaotic patterns. A novel technique to reduce and analyze such complex systems is introduced in this work, and its applications to multiple perturbed systems are discussed comprehensively. In this work, a unified approach between the Floquet theory for time periodic systems and the Poincare theory of Normal Forms is proposed to analyze time varying systems. The proposed unified approach is initially verified for linear time periodic systems with the aid of an intuitive state augmentation and the method of Time Independent Normal Forms (TINF). This approach also resulted in the closed form expressions for the State Transition Matrix (STM) and Lyapunov-Floquet (L-F) transformation for linear time periodic systems. The application of theory towards stability analysis is further demonstrated with the system of Suction Stabilized Floating (SSF) platform. Additionally, multiple control strategies are discussed and implemented to drive an unstable time periodic system to a desired stable point or orbit efficiently and optimally. The computed L-F transformation is further utilized to analyze nonlinear and externally excited systems with deterministic and stochastic time periodic coefficients. The central theme of this work is to verify the extension of Floquet theory towards time varying systems with periodic coefficients comprising of incommensurate frequencies or quasi-periodic systems. As per Floquet theory, a Lyapunov-Perron (L-P) transformation converts a time-varying quasi-periodic system to a time-invariant form. A class of commutative quasi-periodic systems is introduced to demonstrate the proposed theory and its applications analytically. An extension of the proposed unified approach towards analyzing the linear quasi-periodic system is observed to provide good results, computationally less complex and widely applicable for strongly excited systems. The computed L-P transformation using the unified theory is applied to analyze both commutative and non-commutative linear quasi-periodic systems with nonlinear terms and external excitation terms. For highly nonlinear quasi-periodic systems, the implementation of multiple order reduction techniques and their performance comparisons are illustrated in this work. Finally, the robustness and stability analysis of nonlinearly perturbed and stochastically excited quasi-periodic systems are performed using Lyapunov's direct method and Infante's approach.

Touch plays a vital role in maintaining human relationships through social andemotional communications. This research proposes a multi-modal haptic display capable of generating vibrotactile and thermal haptic signals individually and simultaneously. The main objective for creating this device is to explore the importance of touch in social communication, which is absent in traditional communication modes like a phone call or a video call. By studying how humans interpret haptically generated messages, this research aims to create a new communication channel for humans. This novel device will be worn on the user's forearm and has a broad scope of applications such as navigation, social interactions, notifications, health care, and education. The research methods include testing patterns in the vibro-thermal modality while noting its realizability and accuracy. Different patterns can be controlled and generated through an Android application connected to the proposed device via Bluetooth. Experimental results indicate that the patterns SINGLE TAP and HOLD/SQUEEZE were easily identifiable and more relatable to social interactions. In contrast, other patterns like UP-DOWN, DOWN-UP, LEFTRIGHT, LEFT-RIGHT, LEFT-DIAGONAL, and RIGHT-DIAGONAL were less identifiable and less relatable to social interactions. Finally, design modifications are required if complex social patterns are needed to be displayed on the forearm.

This research seeks to present the design and testing of exoskeletons capable of assisting with walking gait, squatting, and fall prevention activities. The dissertation introduces wearable robotics and exoskeletons and then progresses into specific applications and developments in the targeted field. Following the introduction, chapters present and discuss different wearable exoskeletons built to address known issues with workers and individuals with increased risk of fall. The presentation is concluded by an overall analysis of the resulting developments and identifying future work in the field.

The field of prostheses and rehabilitation devices has seen tremendous advancement since the ’90s. However, the control aspect of the said devices is lacking. The need for mathematical theories to improve the control strategies is apparent. This thesis attempts to bridge the gap by introducing some dynamic system analysis and control strategies.Firstly, the human gait dynamics are assumed to be periodic. Lyapunov Floquet theory and Invariant manifold theory are applied. A transformation is obtained onto a simple single degree of freedom oscillator system. The said system is transformed back into the original domain and compared to the original system. The results are discussed and critiqued. Then the technique is applied to the kinematic and kinetic data collected from healthy human subjects to verify the technique’s feasibility. The results show that the technique successfully reconstructed the kinematic and kinetic data. Human gait dynamics are not purely periodic, so a quasi-periodic approach is adopted. Techniques to reduce the order of a quasi-periodic system are studied. Lyapunov-Peron transformation (a surrogate of Lyapunov Floquet transformation for quasi-periodic systems) is studied. The transformed system is easier to control. The inverse of the said transformation is obtained to transform back to the original domain. The application of the techniques to different cases (including externally forced systems) is studied. The reduction of metabolic cost is presented as a viable goal for applying the previously studied control techniques. An experimental protocol is designed and executed to understand periodic assistive forces' effects on human walking gait. Different tether stiffnesses are used to determine the best stiffness for a given subject population. An estimation technique is introduced to obtain the metabolic cost using the center of mass's kinematic data. Lastly, it is concluded that the mathematical techniques can be utilized in a robotic tail-like rehabilitation device. Some possible future research ideas are provided to implement the techniques mentioned in this dissertation.