Browsing by Author "Patel, Amir"

Now showing 1 - 14 of 14
  • Thumbnail Image
    Item
    Open Access
    A contact-implicit direct trajectory optimization scheme for the study of legged maneuverability
    (2022) Shield, Stacey; Patel, Amir
    For legged robots to move safely in unpredictable environments, they need to be manoeuvrable, but transient motions such as acceleration, deceleration and turning have been the subject of little research compared to constant-speed gait. They are difficult to study for two reasons: firstly, the way they are executed is highly sensitive to factors such as morphology and traction, and secondly, they can potentially be dangerous, especially when executed rapidly, or from high speeds. These challenges make it an ideal topic for study by simulation, as this allows all variables to be precisely controlled, and puts no human, animal or robotic subjects at risk. Trajectory optimization is a promising method for simulating these manoeuvres, because it allows complete motion trajectories to be generated when neither the input actuation nor the output motion is known. Furthermore, it produces solutions that optimize a given objective, such as minimizing the distance required to stop, or the effort exerted by the actuators throughout the motion. It has consequently become a popular technique for high-level motion planning in robotics, and for studying locomotion in biomechanics. In this dissertation, we present a novel approach to studying motion with trajectory optimization, by viewing it more as “trajectory generation” – a means of generating large quantities of synthetic data that can illuminate the differences between successful and unsuccessful motion strategies when studied in aggregate. One distinctive feature of this approach is the focus on whole-body models, which capture the specific morphology of the subject, rather than the highly-simplified “template” models that are typically used. Another is the use of “contact-implicit” methods, which allow an appropriate footfall sequence to be discovered, rather than requiring that it be defined upfront. Although contact-implicit methods are not novel, they are not widely-used, as they are computationally demanding, and unnecessary when studying comparatively-predictable constant speed locomotion. The second section of this dissertation describes innovations in the formulation of these trajectory optimization problems as nonlinear programming problems (NLPs). This “direct” approach allows these problems to be solved by general-purpose, open-source algorithms, making it accessible to scientists without the specialized applied mathematics knowledge required to solve NLPs. The design of the NLP has a significant impact on the accuracy of the result, the quality of the solution (with respect to the final value of the objective function), and the time required to solve the problem
  • No Thumbnail Available
    Item
    Open Access
    AeroDima: a cheetah-inspired aerodynamic tail for rapid manoeuvrability
    (2024) Bright, Daryn; Patel, Amir
    The cheetah, the fastest land animal has been hypothesised to use its tail for manoeuvrability and has been the source for many bio-inspired robotic tails. However, the use of a lightweight tail to achieve the same effects has not be studied. This paper goes into the study of using a lightweight, aerodynamic tail to increase manoeuvrability on a wheeled platform, AeroDima. This is achieved by studying the aerodynamics of a cheetah's tail in a wind tunnel to develop a quasi-steady state model. A bio-inspired aerodynamic tail was then designed to maximise drag forces. This bio-inspired tail was also studied in a wind tunnel and compared to the cheetah tail's model. The tail required a platform to operate on, so AeroDima was designed and manufactured. The tail, designed to be a 3DOF underactuated system, was designed to test the effects of the tail on high-speed manoeuvres through a roll motion swing of the tail. The system was tested both in simulation and physical experiments. The simulation, developed with MATLAB's Simscape Multibody toolbox, was designed to be a comprehensive model of AeroDima.
  • No Thumbnail Available
    Item
    Open Access
    Control of Rapid Acceleration in a Planar Legged Robot
    (2023) Mailer, Christopher; Patel, Amir; Govender Reuben
    This thesis details the hardware and control design of Kemba: a planar legged robot intended for investigating bounding and explosive, agile manoeuvres. The robot incorporates both pneumatically actuated knees for powerful, compliant, and impact resistant actuation, and proprioceptive electric actuators at the shoulder and hip for high bandwidth torque control and foot placement. Kemba is capable of bounding at up to 1.7m/s with a full flight phase, jumping just under 1mhigh (2.2 times it's nominal leg length), and accelerating from rest into a top speed bound in only 2 strides and under half a second, demonstrating its agility. Stable bounding and acceleration is achieved using a discrete body oscillation stabiliser, and the more dynamic jumping and somersault motions are generated using offline nonlinear trajectory optimisation. The optimal jumping motion was executed on the physical robot while the somersault is currently still limited to simulation. Due to the unique design and actuator combination, contact implicit trajectory optimisation served as a vital tool for motion identification and controller design. In addition to the robot dynamics and unilateral contact constraints, a more tractable pneumatic actuator model was developed which enabled the numerically stiff, discontinuous air dynamics and discrete valve switching to also be incorporated into the trajectory optimisation formulation. Trajectories resulting from optimisation were accurate enough to be implemented directly on the hardware in the case of the jump motion, and also crucially inform the design of the accelerate from rest controller. The results presented in this work indicate that Kemba is a robust and agile platform, well suited for future work in understanding dynamic manoeuvres and optimal control
  • Thumbnail Image
    Item
    Open Access
    Design of a bipedal robot for rapid acceleration and braking manoeuvres
    (2019) Blom, Alexander Francois; Patel, Amir
    Animals in nature are capable of performing rapid acceleration and braking manoeuvres with ease. However, they have been avoided by researchers due to the complexities of this motion. To investigate and test novel control schemes for such motions, a highly agile mechanical robot is required. The aim of this dissertation was to build a bipedal robot to perform optimal rapid acceleration manoeuvres. This focused on investigating existing robots and using the information therein to design, build and test a new bipedal robot with high agility. The author performed a vigorous investigation into existing actuator schemes and leg topologies that promote agility, balancing the numerous trade-off’s such as mass-specific force and proprioception. This led to the selection of a Quasi-Direct Drive transmission with a scissor linkage leg. Legged robots were generally designed around some known motion [1]. However, selecting suitable mechanical parameters for agile motions with a lack of relevant research was challenging. Trajectory optimisation methods were used to generate unique acceleration motions for bipedal models, aiding in the selection of several physical parameters. With this, a detailed design of Baleka was created, prioritising desirable characteristic for rapid motions. Through several design iterations, the outcome was a fully assembled light weight bipedal robot. All the supporting systems required to operate Baleka were designed and set up, including the Real-time control system, relevant sensors and a boom support to keep it planar. A known metric, vertical agility [2], was used to compare Baleka’s agility to existing robots. Furthermore, a Raibert Controller [3] was also tested on the platform to investigate the robustness of the design. Baleka was found to be the most agile bipedal robot, exceeding the agility of humans. It was able to hop higher than all other robots, verifying it’s suitability for rapid acceleration manoeuvres. However, from the repetitive hopping experiments and high impact forces, slight plastic deformation was witnessed in the gearbox drive shafts.
  • Thumbnail Image
    Item
    Open Access
    Kinematic State Estimation using Multiple DGPS/MEMS-IMU Sensors
    (2020) Ku, Do Yeou; Patel, Amir
    Animals have evolved over billions of years and understanding these complex and intertwined systems have potential to advance the technology in the field of sports science, robotics and more. As such, a gait analysis using Motion Capture (MOCAP) technology is the subject of a number of research and development projects aimed at obtaining quantitative measurements. Existing MOCAP technology has limited the majority of studies to the analysis of the steady-state locomotion in a controlled (indoor) laboratory environment. MOCAP systems such as the optical, non-optical acoustic and non-optical magnetic MOCAP systems require predefined capture volume and controlled environmental conditions whilst the non-optical mechanical MOCAP system impedes the motion of the subject. Although the non-optical inertial MOCAP system allows MOCAP in an outdoor environment, it suffers from measurement noise and drift and lacks global trajectory information. The accuracy of these MOCAP systems are known to decrease during the tracking of the transient locomotion. Quantifying the manoeuvrability of animals in their natural habitat to answer the question “Why are animals so manoeuvrable?” remains a challenge. This research aims to develop an outdoor MOCAP system that will allow tracking of the steady-state as well as the transient locomotion of an animal in its natural habitat outside a controlled laboratory condition. A number of researchers have developed novel MOCAP systems with the same aim of creating an outdoor MOCAP system that is aimed at tracking the motion outside a controlled laboratory (indoor) environment with unlimited capture volume. These novel MOCAP systems are either not validated against the commercial MOCAP systems or do not have comparable sub-millimetre accuracy as the commercial MOCAP systems. The developed DGPS/MEMS-IMU multi-receiver fusion MOCAP system was assessed to have global trajectory accuracy of _0:0394m, relative limb position accuracy of _0:006497m. To conclude the research, several recommendations are made to improve the developed MOCAP system and to prepare for a field-testing with a wild animal from a family of a terrestrial megafauna.
  • Thumbnail Image
    Item
    Open Access
    Markerless 3D Motion Capture of Cheetahs in the Wild
    (2021) Clark, Liam James; Patel, Amir
    3D motion-capture is essential for research in biomechanics and biomedical engineering. It can provide insight into performance and injury prevention in sport, diagnosis of illnesses and disorders as well as help biologists to study animals and help engineers to design bio-inspired robots. Researchers studying animal locomotion often study humans and dogs due to the difficulty associated with studying other animals; however, the cheetah (Acinonyx jubatus) is a particularly compelling animal to study due to its various cursorial adaptions. In recent years, there have been significant improvements in computer-based pattern recognition in deep learning, specifically convolutional neural networks (CNNs). This project will explore the use of computer vision techniques including CNNs, extended Kalman filters (EKFs), non-linear optimisation and sparse bundle adjustment (SBA) to remove the need for markers to be used in recovering a subject's location from video footage. The result of the project will be the development of a markerless 3D motion-capture system. The thesis discusses the theory behind and describes the development of tools for video synchronisation and processing, camera calibration, pose estimation, robust 3D reconstruction and 3D pose visualisation. Results are shown for motion capture performed on video footage of cheetahs. Visualisations and 3D motion data of these agile animals are also shown. The system developed is an enabling tool in the study of biomechanics and biomedical research.
  • Thumbnail Image
    Item
    Open Access
    Optimal Control of the Cheetah During Rapid Manoeuvres
    (2022) Knemeyer, Alexander; Patel, Amir
    Cheetahs are incredibly fast, manoeuvrable and highly dynamic, but relatively little is understood about how this is achieved. Thus, understanding their abilities is a subject of research for roboticists and biologists. Trajectory optimisation is a tool often used to increase our understanding of cheetahs, but current approaches which handle the full complexity of poorly understood manoeuvres are slow. The lack of data means that there are no simulated models of cheetahs known to be representative of dynamic movements such as acceleration and turning. In this project, a modelling change is investigated that decreases the time to find trajectories for models involving long serial chains of rigid bodies. Leveraging this development, a software library is created which facilitates the process of finding trajectories of models of legged robots and animals. Using this library, a complex model of a cheetah is developed, based on real data and some experimentation. Finally, the model is used to generate high speed dynamic manoeuvres which present progress towards understanding the incredible abilities of cheetahs.
  • Thumbnail Image
    Item
    Open Access
    Rapid acceleration of legged robots: a pneumatic approach
    (2021) Van Zyl, Joshua; Patel, Amir; Govender, Reuben
    For robotics to be useful to the public in a multifaceted manner, they need to be both legged and agile. The legged constraint arises as many environments and systems in our world are tailored to ablebodied adults. Therefore, a practically useful robot would need to have the same morphology for maximum efficacy. For robots to be useful in these environments, they need to perform at least as well as humans, therefore presenting the agility constraint. These requirements have been out of reach of the field until recently. The aim of this thesis was to design a planar monopod robot for rapid acceleration manoeuvres, that could later be expanded to a planar quadruped robot. This was achieved through a hybrid electric and pneumatic actuation system. To this end, modelling schemes for the pneumatic cylinder were investigated and verified with physical experiments. This was done to develop accurate models of the pneumatic system that were later used in simulation to aid in the design of the platform. The design of the platform was aided through the use of Simulink to conduct iterative testing and multivariate evaluations using Monte Carlo simulation methods. Once the topology of the leg was set, the detail design was conducted in Solidworks and validated with its built in simulation functions. In addition to the mechanical design of the platform, a specialist boom was designed. The design needed to compensate for the forces the robot exerts on the boom as well as the material constraints on the boom. This resulted in the development of a cable-stayed, four bar mechanism boom system. An embedded operating system was created to control the robot and take in and fuse sensor inputs. This was run using multiple sensors, sub-controllers and microcontrollers. Sensor fusion for the system was done using a Kalman Filter to improve readings and estimate unmeasured states of the robot. This Kalman Filter took LiDAR and accelerometer readings as inputs to the system to produce a subcentimetre accurate position measure for the system. Finally, the completed platform was validated using fixed-body forward hopping tests. These tests showed a significant degree of similarity to the simulated results and therefore validated the design process.
  • Thumbnail Image
    Item
    Open Access
    State estimation of a cheetah spine and tail using an inertial sensor network
    (2015) Fisher, Callen; Patel, Amir; Boje, Edward
    The cheetah (Acinonyx jubatus) is by far the most manoeuvrable and agile terrestrial animal. Little is known, in terms of biomechanics, about how it achieves these incredible feats of manoeuvrability. The transient motions of the cheetah all involve rapid flicking of its tail and flexing of its spine. The aim of the research was to develop tools (hardware and software) that can be used to gain a better understanding of the cheetah tail and spine by capturing its motion. A mechanical rig was used to simulate the tail and spine motion. This insight may inspire and aid in the design of bio-inspired robotic platforms. A previous assumption was that the tail is heavy and acts as a counter balance or rudder, yet this was never tested. Contrary to this assumption, necropsy results determined that the tail was in fact light with a relatively low inertia value. Fur from the tail was used in wind tunnel experiments to determine the drag coefficient of a cheetah tail. No researchers have actively sought to track the motion of a cheetah's spine and tail during rapid manoeuvres via placing multiple sensors on a cheetah. This requires the development of a 3D dynamic model of the spine and tail to accurately study the motion of the cheetah. A wireless sensor network was built and three different filters and state estimation algorithms were designed and validated with a mechanical rig and camera system. The sensor network consists of three sensors on the tail (base, middle and tip) as well as a hypothetical collar sensor (GPS and WiFi were not implemented).
  • Thumbnail Image
    Item
    Open Access
    The Force Floor: Design and Development of a Low-Cost 3D Force Sensing Area Which Utilises Machine Learning to Estimate 3D GRF and CoP from Single-Axis Loadcells
    (2023) Stickells, Devin; Patel, Amir; Albertus Yumna
    Along with motion capture tools, ground reaction force (GRF) sensors form the crux of objective biomechanical analysis. Advances in computer vision have significantly lowered the costs associated with 3D motion capture, but the same cannot be said of 3-axis force plates – the gold standard for GRF capture. If wholistic biomechanics analysis is to become more accessible, a more affordable method of 3D GRF measurement is needed. Single-axis loadcells are significantly cheaper than their 3-axis equivalents, though when axes are not mechanically isolated there is the possibility for crosstalk and the absorption of forces which cannot be measured, leading to a system that cannot be fully described analytically - and is possibly nonlinear in its behaviour. This research investigates the design and small-scale manufacture (to 20 units) of a low-cost force plate design that utilises a machine learning model to overcome these limitations and estimate 3D GRF and centre of pressure from a series of single-axis loadcells. A literature review was performed to understand and compare the relevant approaches to the core aspects of the project. An early proof of concept plate was built and tested along with a simple neural network to establish the feasibility of the idea. Following further investigation, it was discovered that the internal geometry of the plate played an integral role in its accuracy. To this end, the force plate was simulated, and an extensive hardware design process undertaken prior to the design of a full-scale prototype. It was subsequently hypothesised that the ease of repetition of the design could be aided by the development of an automated data creation rig, as well as the use of recently-developed machine learning techniques which reduce data dependency, such as Sim2Real transfer learning and physicsinformed residual networks. A data creation rig was built for purpose. Twenty prototype plates were built, with sixteen of them being interlinked to create the prototype Force Floor - a large force sensing area. The performance of a subset of these plates and their corresponding models was tested against an Advanced Mechanical Technology Inc. (AMTI) BMS6001200 force plate, with the best obtaining average measurement disagreements in the X-, Y- and Z-directions of 1.23, 1.08, and 1.11 percent of the full-scale force respectively (with full-scale deflections of 600 N, 600 N and 2000 N respectively). Analysis of the project's results was encouraging as far as the viability of this design and approach for use in the production of an affordable 3-axis force plate is concerned.
  • Thumbnail Image
    Item
    Open Access
    Trajectory Optimisation Inspired Design for Legged Robotics
    (2021) Fisher, Callen; Patel, Amir
    Control of legged robots is a non-trivial task, especially when looking at aperiodic (non-steadystate) manoeuvres such as rapid acceleration and deceleration. Observing nature, animals are seen to effortlessly perform these rapid transient manoeuvres, however, robust walking is still considered a complex task in the robotics literature. For robots to successfully explore the unknown world outside of the laboratory environment, these transient manoeuvres need to be thoroughly understood and mastered. As controlling and designing legged robots is an extremely complex task, this thesis argues that trajectory optimisation methods can be employed to make various aspects of design more tractable. First trajectory optimisation methods were employed to determine how legged robots should accelerate. In nature animals are seen to leap straight into the desired gait, however, in the robotics literature, robots often perform multiple discrete gait transitions, or accelerate extremely slowly. The results of the study revealed that the optimal method to accelerate was to launch straight into the desired gait (for both bipeds and quadrupeds), with a sliding mass template model emerging for all results. Another discrepancy between the literature and nature is that animals have active spines which have been shown to aid in rapid locomotion tasks. In the robotics literature a number of spines exists, however, which is the optimal morphology for transient manoeuvres? Using Monte Carlobased trajectory optimisation methods, the rigid, revolute and prismatic spine morphologies were compared, with results showing the prismatic spine was the optimal configuration for long-time-horizon tasks. Due to transient locomotion requiring accurate and complex whole-body models, resulting in computationally expensive optimisation problems, an “optimisation-inspired” approach (akin to “bioinspiration”) was taken to identify heuristics and trends for a monopod robot. Initially, optimal trajectories for a two link leg monopod were analysed. Interestingly, during the stance phase, the axial force of the leg behaved in a “bang-bang” like fashion with fine hip torque control. Furthermore the aerial phase showed correspondence to the popular Raibert style controller. This resulted in the development of a hybrid pneumatic-electric monopod robot as a test-bed. Trajectories were then optimised for the robot to determine if these heuristics held. From these results, a stance phase PD controller for the hip actuator was developed and simulated under disturbances to test robustness. The resulting controller and heuristic was successfully tested on the platform, performing a long-time-horizon motion, which included transient phases of acceleration and deceleration.
  • Thumbnail Image
    Item
    Open Access
    UAV collision avoidance: a specific acceleration matching approach
    (2011) Patel, Amir; Winberg, Simon
    An increased level of autonomy is required for future Unmanned Aerial Vehicle (UAV) missions. One of the technologies required for this to occur is an adequate sense and avoid system to enable the UAV to detect threat aircraft and take evasive action if required. This thesis investigates a collision avoidance system to satisfy a significant portion of the requirements for sense and avoid. It was hypothesised that a recently published method of UAV guidance, Specific Acceleration Matching (SAM) Control, could address the shortcomings of the current implementations. Additionally, a novel algorithm, the Linear 3D Velocity Guidance Control Algorithm (3DVGC) was developed to address the particular requirements of UAV collision avoidance.
  • Thumbnail Image
    Item
    Open Access
    Understanding the motions of the cheetah tail using robotics
    (2015) Patel, Amir; Braae, Martin
    The cheetah is capable of incredible feats of manoeuvrability. But, what is interesting about these manoeuvres is that they involve rapid swinging of the animal's lengthy tail. Despite this, very little is understood about the cheetah tail and its motion, with the common view being that it is "heavy" and possibly used as a "counter balance" or as a "rudder". In this dissertation, this subject is investigated by exploring the motions of the cheetah tail by means of mathematic al models, feedback control and novel robot platforms. Particularly, the motion in the roll axis is first investigated and it is determined that it assists stability of high speed turns. This is validated by modelling and experimental testing on a novel tailed robot, Dima I. Inspired by cheetah video observations, the tail motion in the pitch axis during rapid acceleration and braking manoeuvres is also investigated. Once again modelling and experimental testing on a tailed robot are performed and the tail is shown to stabilise rapid acceleration manoeuvres. Video observations also indicate the tail movement in the shape of a cone: a combination of pitching and yawing. Understanding this motion is done by setting up an optimization problem. Here, the optimal motion was found to be to a cone which results in a continuous torque on the body during a turn while galloping. A novel two degree of freedom tailed robot, Dima II, was then developed to experimentally validate the effect of this motion. Lastly, measurement of the cheetah tail inertia was performed during a routine necropsy where it was found to have lower inertia than assumed. However, the tail has thick, long fur that was tested in a wind tunnel. Here it was found that the furry tail is capable of producing significant drag forces without a weight penalty. Subsequently, mathematical models incorporating the aerodynamics of the tail were developed and these were used to demonstrate its effectiveness during manoeuvres.
  • No Thumbnail Available
    Item
    Open Access
    WildPose: Long-Range 3D Motion Tracking of Cheetahs in the Wild Using Multi-Sensor Fusion
    (2023) Joska, Daniel; Patel, Amir
    In many fields of research, it is often desirable to estimate the 3D pose of a subject - human, animal, or otherwise. Methods for obtaining accurate 3D pose data of a subject are broad in their applications; they inform the design of bio-mimetic robots, they aid greatly in bio-mechanical research, and they are used commonly in the study of animal neuroscience. Currently, robust methods for long-range tracking of subjects in the wild are few and far between, given the rarity of specific training data as well as the generally challenging nature of the associated computer vision problems. This thesis describes the design, implementation, and testing of both a hardware and software component to a method for the 3D motion capture of cheetahs in the wild, dubbed WildPose. The method makes use of multi-sensor fusion including lidar, RGB and IMU sensors to compensate for situations where pure vision-based techniques perform inadequately. To increase robustness, the software design makes use of previously successful trajectory optimisation techniques to yield accurate pose data in adverse conditions that would otherwise be extremely difficult to obtain. The method is extendable to other species with minimal variations. We demonstrate the method's efficacy through experimental validation on challenging cheetah locomotion datasets collected in the wild, presenting both qualitative and quantitative analyses for varied movements, environments, and lighting conditions. Through the shown effectiveness of these techniques in our specific use case, we aim to prove that the methods used perform admirably even under the trickiest of reconstruction problems. Thereby, we present our findings on cheetahs as a promising blueprint for the 3D motion capture of other species.