Search

Heightened demand for alternatives to human exposure to strenuous and repetitive labour, as well as to hazardous environments, has led to an increased interest in real-world deployment of robotic agents. Targeted applications require robots to be adept at synthesising complex motions rapidly across a wide range of tasks and environments. To this end, this thesis proposes leveraging abstractions of the problem at hand to ease and speed up the solving. We formalise abstractions to hint relevant robotic behaviour to a family of planning problems, and integrate them tightly into the motion synthesis process to make real-world deployment in complex environments practical. We investigate three principal challenges of this proposition. Firstly, we argue that behavioural samples in form of trajectories are of particular interest to guide robotic motion synthesis. We formalise a framework with behavioural semantic annotation that enables the storage and bootstrap of sets of problem-relevant trajectories. Secondly, in the core of this thesis, we study strategies to exploit behavioural samples in task instantiations that differ significantly from those stored in the framework. We present two novel strategies to efficiently leverage offline-computed problem behavioural samples: (i) online modulation based on geometry-tuned potential fields, and (ii) experience-guided exploration based on trajectory segmentation and malleability. Thirdly, we demonstrate that behavioural hints can be extracted on-the-fly to tackle highlyconstrained, ever-changing complex problems, from which there is no prior knowledge. We propose a multi-layer planner that first solves a simplified version of the problem at hand, to then inform the search for a solution in the constrained space. Our contributions on efficient motion synthesis via behaviour guidance augment the robots' capabilities to deal with more complex planning problems, and do so more effectively than related approaches in the literature by computing better quality paths in lower response time. We demonstrate our contributions, in both laboratory experiments and field trials, on a spectrum of planning problems and robotic platforms ranging from high-dimensional humanoids and robotic arms with a focus on autonomous manipulation in resembling environments, to high-dimensional kinematic motion planning with a focus on autonomous safe navigation in unknown environments. While this thesis was motivated by challenges on motion synthesis, we have explored the applicability of our findings on disparate robotic fields, such as grasp and task planning. We have made some of our contributions open-source hoping they will be of use to the robotics community at large.

In robotics, robustness is an important and desirable attribute of any system, from perception to planning and control. Robotic systems need to handle numerous factors of uncertainty when they are deployed, and the more robust a method is, the fewer chances there are of something going wrong. In planning and control, being robust is crucial to deal with uncertain contact timings and positions, mismatches in the dynamics model of the system, noise in the sensor readings and communication delays. In this thesis, we focus on the problem of dealing with uncertainty and external disturbances applied to the robot. Reactive robustness can be achieved at the control stage using a variety of control schemes. For example, model predictive control approaches are robust against external disturbances thanks to the online high-frequency replanning of the motion being executed. However, taking robustness into account in a proactive way, i.e., during the planning stage itself, enables the adoption of kinematic configurations that allow the system as a whole to better deal with uncertainty and disturbances. To this end, we propose a novel trajectory optimisation framework for robotic systems, ranging from fixed-base manipulators to legged robots, such as humanoids or quadrupeds equipped with arms. We tackle the problem from a first-principles perspective, and define a robustness metric based on the robot's capabilities, such as the torques available to the system (considering actuator torque limits) and contact stability constraints. We compare our results with other existing approaches and, through simulation and experiments on the real robot, we show that our method is able to plan trajectories that are more robust against external disturbances.

Robotic systems, of various sizes and types, are not only becoming more prominent in the research communities but have been further pushed into industrial and business uses. With the advent of more capable hardware able to run longer and withstand these industrial environments, we can start really exploring the uses of these robots in their given workspaces. To have robots performing in these areas, we require the ability to control them on a low level, enabling consistent and successful motion, before being able to achieve more complex tasks with them. Whilst heavy research has been done for low level control typically these rely heavily on a model of the robot. In particular, in the cases where we wish to control the forces on the robot, we require an accurate model of these forces caused by various physical effects on the robot, such as inertia and gravity. Obtaining these models has been a researched topic for many years with various approaches being discussed and tested. These approaches range from: obtaining the model parameters from computer aided design software, to data driven approaches that use the likes of neural networks to learn the mapping from the robot state to the robot forces, to adaptive control frameworks that attempt to learn the model of the dynamics whilst providing stability guarantees. In this thesis, we initially look deeper into the various approaches taken by previous researchers in the field of dynamics learning. In particular we delve into the advantage of Semi-Parametric modelling techniques on robots and we hypothesize the expansion of Semi-Parametric models to adaptive control frameworks allowing for simultaneous control and learning, with potential stability guarantees. Furthermore we predict that we should be able to expand this idea of adaptive control to more easily fit complex constrained robots, such as quadrupeds. In order to prove and support these hypothesis' we look at each one individually, starting with a deeper analysis into the previous research. With this analysis, we identified a lack of software support for adaptive control which would greatly slow down future progress in the field. Hence we introduced the ARDL library, which has been specifically designed to support our and others work in adaptive control algorithms. We demonstrate the effectiveness of the library through the implementation of various adaptive controllers on a simulated manipulator. In each implementation we are able to show the minimization of tracking and torque errors as appropriate to the controller. ARDL is then used to support the development of the resulting chapters. We then shift our focus onto exploring Semi-Parametric models. We look specifically into creating an online Semi-Parametric model that allows us to control and learn the model simultaneously. We use a mix of a composite adaptive control algorithm to train the Parametric component and use a Gaussian Mixture model as the Non-Parametric component that is trained through an iterative algorithm. We identify a key issue that arises when training both models simultaneously. Which demonstrates an inconsistency between the two models when both are updated simultaneously. We demonstrate the instability that arises without compensating for the inconsistency using a simulated robot. We then identify a general corrective term that we specialize for the Gaussian Mixture Model. By using this new transform we demonstrate, for the second joint, that we go from an unstable behaviour, that generates tracking errors up to +/- 0.2 rad in position and +/- 0.35 rad/s in velocity which increases over time, to a more stable behaviour that reduces and maintains the tracking error to between +/-0.05 rad in position and +/-0.1 rad/s in velocity. We also provide the nMSE results for the whole trajectory for every joint. The result over every metric shows that over every joint we perform better with some metrics and joints having improvements of up to a factor of 100 such as the second joint torque predictions going from an nMSE of 1 to 0.01. We extend the consistency transform framework to a Radial Basis Function Neural Network to further emphasise the generalisability of the consistency correction. In a similar fashion to the Gaussian Mixture model, we are able to update the Radial Basis Function Neural Network through a gradient decent based algorithm. Using this gradient based update we are able to define a consistency transform computed through a gradient update. By updating this transform, alongside the main model, at any step we know the correct transform to adjust the model to a change in Parametric inertial parameters. Using the transform we are able to reconfirm our results with the Gaussian Mixture model such that the Radial Basis Function Neural Network demonstrates very similar online performance on a simulated Kuka LWR IV. In particular our nMSE for the second joint drops down to +/-2.052 x10-7rad in position, +/-6.959 x10-7rad s-1 in velocity, and +/-7.573 x10-7Nm in torque after learning. We also initially explore a rough stability proof for the algorithm, demonstrating where the consistency transform cancels out the inconsistency arising during learning. Finally we look into extending the model learning into a constrained floating base system. Whilst some research has been achieved on holonomically constrained systems, we implement an adaptive controller that needs minimal manual computation. We focus on taking the implementation of the direct adaptive control and extending it to the anymal quadruped platform, and an underactuated two link planar robot. We use constraint projections and constraint consistent errors to achieve this control. Thus, we are able to prove the Lyapunov stability of the task-space target. We then demonstrate the adaptive control on the simulated anymal and planar robot, and the results demonstrate a clear reduction of error, whilst adapting the model in a similar fashion to the direct adaptive control on a fully actuated robot. We specifically drop down to an error of 0.01m and 0.01rad for the pose, and 0:01m/s and 0.01rad/s for the Cartesian velocities whilst learning the inertial parameters.

The main goal of this thesis is to accomplish a compliant and stable closed-loop physical human-robot cooperation by guaranteeing the safety metrics for all of the agents in a shared-working environment. There is increasing interest in control frameworks capable of moving robots from industrial cages to unstructured environments and coexisting with humans. Initially, having robots capable of safely interacting with humans was of interest for medical applications (e.g., rehabilitation, surgical). Despite significant improvement in some specific applications like medical robotics, there is still the need for a general control framework that improves interaction robustness and motion dynamics. Passive controllers show promising results in this direction; however, they often rely on virtual energy tanks that can guarantee passivity as long as they do not run out of energy. In this thesis, a fractal attractor is proposed to implement a variable impedance controller that can retain passivity without relying on energy tanks. The controller generates a fractal attractor around the desired state using an asymptotic stable potential field, making the controller robust to discretization and numerical integration errors. Thus, the proposed Fractal Impedance Controller (FIC) in this thesis is robust for low-bandwidth applications. I have tested this controller with a torque controlled 7-DoF manipulator. The results prove that it can accurately track both trajectories and end-effector forces during interaction. Furthermore, it can automatically deal with the extra energy introduced by changes in interaction conditions, null-space controller and environment. Therefore, on the one hand these properties make the controller ideal for applications where the dynamic interaction at the end-effector is challenging to be characterized a priori, such as proximate physical human-robot cooperation and unknown dynamics. On the other hand in remote human-robot cooperation, robotic teleoperation provides human-in-the-loop capabilities of complex manipulation tasks in dangerous or remote environments, such as planetary exploration or nuclear decommissioning. This thesis proposes a novel bilateral telemanipulation architecture using the proposed passive FIC, which does not depend upon an active viscous component for guaranteeing stability. Compared to a traditional impedance controller in ideal conditions (no delays and maximum communication bandwidth), the proposed method yields higher transparency in interaction and demonstrates superior dexterity and capability in my telemanipulation test scenarios. I also validate its performance with extreme delays up to 1s and communication bandwidths as low as 10Hz. The results of the carried out experiments validate a consistent stability when using the proposed controller in challenging conditions, regardless of operator expertise. The proposed fractal impedance controller in this thesis exploits its non-linear stiffness to adapt to multiple cooperative scenarios without tuning the controller. Furthermore, the FIC has an intuitive method to adjust the impedance that can be performed online without affecting stability. The experimental results, carried out using 2 torque controlled 7-DoF manipulators and the Sigma.7 haptic device, also show that the proposed method can perform tasks such as drilling, moving objects with unknown dynamics, and interacting with humans without re-tuning the controller's impedance in a tele-cooperative manner consisting of multi-agents in the loop. The FIC also allows identifying the highest impedance profile for a robot experimentally, and it bounds the maximum momentum generated while moving. Thus, it opens new possibilities for developing better adaptive controllers by coupling the proposed method with learning and optimisation algorithms to modulate its behaviour without the risk of incurring instability issues.

Assistive robotic devices such as exoskeletons and prosthetic limbs have great potential as tools for both augmentation and rehabilitation. However, due to the complexity of controlling these devices, especially in unstructured environments where factors such as walking speed and incline can vary rapidly, it is uncommon to see exoskeletons outside of a clinical or research setting. Prostheses, whilst more common, are typically passive, which limits their ability to match the push off forces associated with healthy gait. Motivated by modern techniques for controlling legged robots, this thesis motivates the pursuit of an optimisation-based approach to the control and design of exoskeletons. We identify a number of open problems within the field, namely (1) how to model the dynamic interaction between a human subject and an attached exoskeleton; (2) identifying the appropriate metric or combination of metrics to optimise for in exoskeleton-assisted locomotion; and (3) how to account for changes in human walking style induced by the presence of external assistive forces. This thesis details attempts to solve each of these problems. We present a methodology for expressing human-exoskeleton system models as a combination of musculoskeletal models, exoskeleton inertial parameters and constraint forces. A specific human-exoskeleton model is detailed, along with a range of methods for modelling the interaction forces which occur at the attachment points between the human and exoskeleton agents. Experimental motion data is analysed using musculoskeletal modelling software (OpenSim) to quantify the effect that each of these interaction models, which represent various degrees of approximation, have on the resulting humanexoskeleton dynamics. Applying exoskeleton assistance is inherently a shared control problem. The overall goal is not to achieve a prescribed motion at any cost, or to do so while minimising exoskeleton joint torques, but rather to enhance aspects of the assisted humans motions; for example, increasing energy efficiency or stability. Therefore, in order to optimise exoskeleton control patterns we must first consider what it means for the resultant gait patterns to be optimal, or even good. We present a detailed analysis of exoskeleton-assisted walking in healthy subjects, with a particular focus on identifying those metrics which are invariant to changes in walking condition (e.g. walking speed or incline). We posit that such metrics, which exhibit strong invariance properties, are good candidates for the objective function of an optimisation-based controller. Human walking strategies are unique and complex, and the problem of predicting the effect of exoskeleton assistance on a subjects gait pattern is a challenging one. In recent years, success has been had by methods which aim to learn suitable assistance strategies directly from a subject, via a process known as human-in-the-loop optimisation. We present a novel humanin- the-loop framework which utilises musculoskeletal modelling to make the learning process more time-efficient. Our method is evaluated on a number of subjects walking on a treadmill with exoskeleton assistance. In addition, we also explore how human-in-the-loop optimisation can be used to inform the design of exoskeletons to enhance their assistive capabilities. Overall, these contributions represent a step towards enabling the wider usage of exoskeletons and other assistive robotic devices, which could lead to significant improvements to quality of life for many.

A significant number of Machine Learning methods for automation currently rely on data-hungry training techniques. The lack of accessible training data often represents an insurmountable obstacle, especially in the fields of robotics and automation, where acquiring new data can be far from trivial. Additional data acquisition is not only often expensive and time-consuming, but occasionally is not even an option. Furthermore, the real world applications sometimes have commercial sensitivity issues associated with the distribution of the raw data. This doctoral thesis explores bypassing the aforementioned difficulties by synthesising new realistic and diverse datasets using the Generative Adversarial Network (GAN). The success of this approach is demonstrated empirically through solving a variety of case-specific data-hungry problems, via application of novel GAN-based techniques and architectures. Specifically, it starts with exploring the use of GANs for the realistic simulation of the extremely high-dimensional underwater acoustic imagery for the purpose of training both teleoperators and autonomous target recognition systems. We have developed a method capable of generating realistic sonar data of any chosen dimension by image-translation GANs with Markov principle. Following this, we apply GAN-based models to robot behavioural repertoire generation, that enables a robot manipulator to successfully overcome unforeseen impedances, such as unknown sets of obstacles and random broken joints scenarios. Finally, we consider dynamical system identification for articulated robot arms. We show how using diversity-driven GAN models to generate exploratory trajectories can allow dynamic parameters to be identified more efficiently and accurately than with conventional optimisation approaches. Together, these results show that GANs have the potential to benefit a variety of robotics learning problems where training data is currently a bottleneck.

Optimal loco-manipulation planning and control for high-dimensional systems based on general, non-linear optimisation allows for the specification of versatile motion subject to complex constraints. However, complex, non-linear system and environment dynamics, switching contacts, and collision avoidance in cluttered environments introduce non-convexity and discontinuity in the optimisation space. This renders finding optimal solutions in complex and changing environments an open and challenging problem in robotics. Global optimisation methods can take a prohibitively long time to converge. Slow convergence makes them unsuitable for live deployment and online re-planning of motion policies in response to changes in the task or environment. Local optimisation techniques, in contrast, converge fast within the basin of attraction of a minimum but may not converge at all without a good initial guess as they can easily get stuck in local minima. Local methods are, therefore, a suitable choice provided we can supply a good initial guess. If a similarity between problems can be found and exploited, a memory of optimal solutions can be computed and compressed efficiently in an offline computation process. During runtime, we can query this memory to bootstrap motion synthesis by providing a good initial seed to the local optimisation solver. In order to realise such a system, we need to address several connected problems and questions: First, the formulation of the optimisation problem (and its parametrisation to allow solutions to transfer to new scenarios), and related, the type and granularity of user input, along with a strategy for recovery and feedback in case of unexpected changes or failure. Second, a sampling strategy during the database/memory generation that explores the parameter space efficiently without resorting to exhaustive measures---i.e., to balance storage size/memory with online runtime to adapt/repair the initial guess. Third, the question of how to represent the problem and environment to parametrise, compute, store, retrieve, and exploit the memory efficiently during pre-computation and runtime. One strategy to make the problem computationally tractable is to decompose planning into a series of sequential sub-problems, e.g., contact-before-motion approaches which sequentially perform goal state planning, contact planning, motion planning, and encoding. Here, subsequent stages operate within the null-space of the constraints of the prior problem, such as the contact mode or sequence. This doctoral thesis follows this line of work. It investigates general optimisation-based formulations for motion synthesis along with a strategy for exploration, encoding, and exploitation of a versatile memory-of-motion for providing an initial guess to optimisation solvers. In particular, we focus on manipulation in complex environments with high-dimensional robot systems such as humanoids and mobile manipulators. The first part of this thesis focuses on collision-free motion generation to reliably generate motions. We present a general, collision-free inverse kinematics method using a combination of gradient-based local optimisation with random/evolution strategy restarting to achieve high success rates and avoid local minima. We use formulations for discrete collision avoidance and introduce a novel, computationally fast continuous collision avoidance objective based on conservative advancement and harmonic potential fields. Using this, we can synthesise continuous-time collision-free motion plans in the presence of moving obstacles. It further enables to discretise trajectories with fewer waypoints, which in turn considerably reduces the optimisation problem complexity, and thus, time to solve. The second part focuses on problem representations and exploration. We first introduce an efficient solution encoding for trajectory library-based approaches. This representation, paired with an accompanying exploration strategy for offline pre-computation, permits the application of inexpensive distance metrics during runtime. We demonstrate how our method efficiently re-uses trajectory samples, increases planning success rates, and reduces planning time while being highly memory-efficient. We subsequently present a method to explore the topological features of the solution space using tools from computational homology. This enables us to cluster solutions according to their inherent structure which increases the success of warm-starting for problems with discontinuities and multi-modality. The third part focuses on real-world deployment in laboratory and field experiments as well as incorporating user input. We present a framework for robust shared autonomy with a focus on continuous scene monitoring for assured safety. This framework further supports interactive adjustment of autonomy levels from fully teleoperated to automatic execution of stored behaviour sequences. Finally, we present sensing and control for the integration and embodiment of the presented methodology in high-dimensional real-world platforms used in laboratory experiments and real-world deployment. We validate our presented methods using hardware experiments on a variety of robot platforms demonstrating generalisation to other robots and environments.