abstract: Most planning agents assume complete knowledge of the domain, which may not be the case in scenarios where certain domain knowledge is missing. This problem could be due to design flaws or arise from domain ramifications or qualifications. In such cases, planning algorithms could produce highly undesirable behaviors. Planning with incomplete domain knowledge is more challenging than partial observability in the sense that the planning agent is unaware of the existence of such knowledge, in contrast to it being just unobservable or partially observable. That is the difference between known unknowns and unknown unknowns. In this thesis, I introduce and formulate this as the problem of Domain Concretization, which is inverse to domain abstraction studied extensively before. Furthermore, I present a solution that starts from the incomplete domain model provided to the agent by the designer and uses teacher traces from human users to determine the candidate model set under a minimalistic model assumption. A robust plan is then generated for the maximum probability of success under the set of candidate models. In addition to a standard search formulation in the model-space, I propose a sample-based search method and also an online version of it to improve search time. The solution presented has been evaluated on various International Planning Competition domains where incompleteness was introduced by deleting certain predicates from the complete domain model. The solution is also tested in a robot simulation domain to illustrate its effectiveness in handling incomplete domain knowledge. The results show that the plan generated by the algorithm increases the plan success rate without impacting action cost too much. / Dissertation/Thesis / Masters Thesis Computer Science 2020

This master’s thesis describes equations of motion for dynamic model of nonholonomic constrained system, namely the trident robotic snakes. The model is studied in the form of Lagrange's equations and D’Alembert’s principle is applied. Actually this thesis is a continuation of the study going at VUT about the simulations of non-holonomic mechanisms, specifically robotic snakes. The kinematics model was well-examined in the work of of Byrtus, Roman and Vechetová, Jana. So here we provide equations of motion and address the motion planning problem regarding dynamics of the trident snake equipped with active joints through basic examples and propose a feedback linearization algorithm.

This thesis deals about design and realization software application "Mobile robot studio" for planning path mobile robot in pseudo 3D world. It contains several tools, witch most important are: simulation control, path planning, world editor and commands editor for CAN. Application was made by technology .NET 2.0 and for 3D design was used Microsoft DirectX 9 API. This thesis has been supported by the Czech Ministry of Education in the frame of MSM 0021630529 Research Intention Inteligent Systems in Automation.

No description available.

Research about unmanned ground vehicles (UGVs) has received an increased amount of attention in recent years, partly due to the many applications of UGVs in areas where it is inconvenient or impossible to have human operators, such as in mines or urban search and rescue. Two closely linked problems that arise when developing such vehicles are motion planning and control of the UGV. This thesis explores these subjects for a UGV with an unknown, and possibly time-variant, dynamical model. A framework is developed that includes three components: a machine learning algorithm to estimate the unknown dynamical model of the UGV, a motion planner that plans a feasible path for the vehicle and a controller making the UGV follow the planned path. The motion planner used in the framework is a lattice-based planner based on input sampling. It uses a dynamical model of the UGV together with motion primitives, defined as a sequence of states and control signals, which are concatenated online in order to plan a feasible path between states. Furthermore, the controller that makes the vehicle follow this path is a model predictive control (MPC) controller, capable of taking the time-varying dynamics of the UGV into account as well as imposing constraints on the states and control signals. Since the dynamical model is unknown, the machine learning algorithm Bayesian linear regression (BLR) is used to continuously estimate the model parameters online during a run. The parameter estimates are then used by the MPC controller and the motion planner in order to improve the performance of the UGV. The performance of the proposed motion planning and control framework is evaluated by conducting a series of experiments in a simulation study. Two different simulation environments, containing obstacles, are used in the framework to simulate the UGV, where the performance measures considered are the deviation from the planned path, the average velocity of the UGV and the time to plan the path. The simulations are either performed with a time-invariant model, or a model where the parameters change during the run. The results show that the performance is improved when combining the motion planner and the MPC controller with the estimated model parameters from the BLR algorithm. With an improved model, the vehicle is capable of maintaining a higher average velocity, meaning that the plan can be executed faster. Furthermore, it can also track the path more precisely compared to when using a less accurate model, which is crucial in an environment with many obstacles. Finally, the use of the BLR algorithm to continuously estimate the model parameters allows the vehicle to adapt to changes in its model. This makes it possible for the UGV to stay operational in cases of, e.g., actuator malfunctions.

Les algorithmes probabilistes offrent de puissantes possibilités quant à la résolution de problèmes de planification de mouvements pour des robots complexes dans des environnements quelconques. Cependant, la qualité des chemins solutions obtenus est discutable. Cette thèse propose un outil pour optimiser ces chemins et en améliorer la qualité. La méthode se base sur l'optimisation numérique contrainte et la détection de collision pour réduire la longueur du chemin tout en évitant les collisions. La modularité des méthodes probabilistes nous a aussi inspirés pour réaliser un algorithme de génération de sauts pour des personnages. Cet algorithme est décrit par trois étapes de planifications, de la trajectoire du centre du personnage jusqu'à son mouvement corps-complet. Chaque étape bénéficie de la rigueur de la planification pour éviter les collisions et pour contraindre le chemin. Nous avons proposé des contraintes inspirées de la physique pour améliorer la plausibilité des mouvements, telles que du non-glissement, de la limitation de vitesse et du maintien de contacts. Les travaux de cette thèse ont été intégrés dans le logiciel "Humanoid Path Planner" et les rendus visuels effectués avec Blender. / Probabilistic algorithms offer powerful possibilities as for solving motion planning problems for complex robots in arbitrary environments. However, the quality of obtained solution paths is questionable. This thesis presents a tool to optimize these paths and improve their quality. The method is based on constrained numerical optimization and on collision checking to reduce the path length while avoiding collisions. The modularity of probabilistic methods also inspired us to design a motion generation algorithm for jumping characters. This algorithm is described by three steps of motion planning, from the trajectory of the character's center to the wholebody motion. Each step benefits from the rigor of motion planning to avoid collisions and to constraint the path. We proposed physics-inspired constraints to increase the plausibility of motions, such as slipping avoidance, velocity limitation and contact maintaining. The thesis works have been implemented in the software `Humanoid Path Planner' and the graphical renderings have been done with Blender.

Autonomous Unmanned Aerial Vehicles (UAV) havegreat potential in executing various complex tasks due to theirflexibility and relatively small size. The aim of this paper is todevelop a motion planner capable of generating a trajectory withaggressive maneuvers through narrow spaces without collision.The approach utilizes a framework using an optimized variantof the Rapidly-exploring Random Tree (RRT) algorithm, calledRRT*, with a Control Barrier Functions (CBF) based obstacleavoidance algorithm as well as a motion primitive generator. If amotion primitive collides with an obstacle, the obstacle avoidancealgorithm will attempt to reach the end state of a motion primitivein a collision free manner while complying with the actuationconstraints. From the collision free trajectories an optimal path iscontinuously searched for by RRT* by minimizing a cost in jerk.The performance of RRT* and the obstacle avoidance are testedin simulations independently and jointly, in several differentscenarios. The resulting motion planner successfully finds ahigh-level trajectory for the different scenarios. Limitations ofthe method as well as possible areas of improvements are alsodiscussed at the end of this paper. / Autonoma UAV har goda möjligheter för att utföra flera olika komplexa uppgifter tack vare deras flexibilitet och storlek. Denna rapport redogör för en rörelseplaneringsalgoritm som kombinerar manövrerbarheten hos en UAV för att skapa en kollisionsfri bana som innehåller aggressiva manövreringar genom trånga utrymmen. Tillvägagångssättet innefattar att kombinera Rapidly-exploring Random Tree (RRT*) med en algoritm för att undvika hinder baserad på Control Barrier Functions (CBF), samt att låta banan delas upp i segment, så kallade motion primitives, som genereras var för sig. Om en motion primitive kolliderar kommer den hinderundvikande algoritmen göra ett försök att nå dess målposition medan kollision undviks och manövreringsbegränsningarna uppfylls. Med en samling genomförbara motion primitives söker RRT* efter en kontinuerlig bana optimerad med hänsyn till en kostnad i ryck. Prestandan för RRT* och den hinderundvikande algoritmen simuleras både separat och tillsammans. Den resulterande rörelseplaneraren lyckas hitta en genomförbar bana för vardera scenario. Begränsningar av metoden samt potentiella förbättringsområden diskuteras i slutet av denna rapport. / Kandidatexjobb i elektroteknik 2021, KTH, Stockholm

The inherited rotation in a caster wheel allows movement in any direction, but pays at the expense of reaction torques. When implemented in a mobile robot, these forces have a negative impact in its performance. One approach is to restrict rotations on the spot by attaching a filter to the output of the motion planner. However, this formulation compromises the navigation’s completion in critical scenarios, such as parking, taking curves in narrow corridors or navigating at the presence of a high density of obstacles. Therefore, in this thesis we consider the influence of caster wheels in the motion planning stage, commonly presented as local planning. This work proposes a Model Predictive Control (MPC) based local planner that integrates the caster wheel physics into the motion planning stage. A caster wheel aware term is combined with a reference tracking based navigation, which leads to the formulation of the Caster Wheel Aware Local Planner (CWAWLP). Since this method requires knowing the caster wheel’s state and there is no sensor that provides this information, a caster wheel state observer is also formulated. In order to evaluate the impact of the caster wheel aware term, CWAWLP is compared to a Caster Wheel based Agnostic Local Planner (CWAGLP) and a Caster Wheel based Agnostic Planner Local Planner with Path Filter (CWPFLP). After running simulations for three case studies in a virtual framework, two experimental case studies are conducted in an intra-logistics robot. These are evaluated according to the navigation’s quality, motor torque usage and energy consumption. According to the patterns observed in the evaluation, CWAWLP covers a longer distance than CWAGLP wihout decreasing the navigation’s quality. At the same time, its motor torques are similar to the ones of CWPFLP. Therefore, CWAWLP is capable of considering caster wheel physics without sacrificing navigation capabilities. The formulated caster wheel aware term is compatible with any MPC based navigation algorithm and inherits the derivation of an observer capable of estimating caster wheel rotation angles and rolling speeds. Even if the caster wheel awareness has been implemented in a differential driven robot, this approach is also applicable to vehicles with an alternative drivetrain, such as car-like robots. / Den ärvda rotationen i ett hjul möjliggör rörelse i vilken riktning som helst, men fås på bekostnad av reaktionsmoment. När de implementeras i en mobil robot har dessa krafter en negativ inverkan på dess prestanda. Ett tillvägagångssätt är att begränsa rotationer på plats genom att applicera ett filter på rörelseplannerns utgång. Denna formulering komprometterar dock navigeringens slutförande i kritiska scenarier, såsom parkering, kurvor i smala korridorer eller navigering i närheten av höga hinder. Därför beaktar vi i denna avhandling påverkan av hjul på hjulplaneringen, som ofta presenteras som lokal planering. Detta arbete föreslår en Model Predictive Control (MPC) -baserad lokal planerare som integrerar svängbara länkhjuls fysik i rörelseplaneringsstadiet. En kugghjulmedveten term kombineras med en referensspårningsbaserad navigering, vilket leder till formuleringen av Caster Wheel Aware Local Planner (CWAWLP). Eftersom denna metod kräver kunskap om svängbara länkhjuls tillstånd och det inte finns någon sensor som ger denna information, formuleras också en hjulhjulstillståndsobservatör. För att utvärdera effekten av det medvetna begreppet svängbara änkhjul jämförs CWAWLP med en Caster Wheel-baserad Agnostic Local Planner (CWAGLP) och en Caster Wheel-baserad Agnostic Planner Local Planner with Path Filter (CWPFLP). Efter att ha kört simuleringar för tre fallstudier i ett virtuellt ramverk genomförs två experimentella fallstudier i en intra-logistikrobot. Dessa utvärderas enligt navigeringens kvalitet, vridmomentanvändning och energiförbrukning. Enligt de mönster som observerats i utvärderingen når CWAWLP ett längre avstånd än CWAGLP utan att sänka navigeringens kvalitet. Samtidigt liknar motorns vridmoment dem som CWPFLP. Därför kan CWAWLP ta hänsyn till svängbara länkhjuls fysik utan att offra navigationsfunktionerna. Den formulerade medhjulningsmedveten termen är kompatibel med vilken MPC-baserad navigationsalgoritm som helst och ärver härledningen av en observatör som kan uppskatta hjulets rotationsvinklar och rullningshastigheter. Även om hjulhjälpmedvetenheten har implementerats i en differentierad robot, är detta tillvägagångssätt också tillämpligt på fordon med ett alternativt drivsystem, såsom billiknande robotar.

This dissertation focuses on constrained control of two applications: helicopter and ocean current turbines (OCT). A major contribution in the helicopter application is a novel model predictive control (MPC) framework for helicopter shipboard operations in high demanding sea-based conditions. A complex helicopter-ship dynamics interface has been developed as a system of implicit nonlinear ordinary differential equations to capture essential characteristics of the nonlinear helicopter dynamics, the ship dynamics, and the ship airwake interactions. Various airwake models such as Control Equivalent Turbulence Inputs (CETI) model and Computation Fluid Dynamics (CFD) data of the airwake are incorporated in the interface to describe a realistic model of the shipborne helicopter. The feasibility of the MPC design is investigated using two case studies: automatic deck landing during the ship quiescent period in sea state 5, and lateral reposition toward the ship in different wind-over-deck conditions. To improve the overall MPC performance, an updating scheme for the internal model of the MPC is proposed using linearization around operating points. A mixed-integer MPC algorithm is also developed for helicopter precision landing on moving decks. The performance of this control structure is evaluated via numerical simulations of the automatic deck landing in adverse conditions such as landing on up-stroke, and down-stroke moving decks with high energy indices. Kino-dynamic motion planning for coordinated maneuvers to satisfy the helicopter-ship rendezvous conditions is implemented via mixed integer quadratic programming. In the OCT application, the major contribution is that a new idea is leveraged from helicopter blade control by introducing cyclic blade pitch control in OCT. A minimum energy, constrained control method, namely Output Variance Constrained (OVC) control is studied for OCT flight control in the presence of external disturbances. The minimum achievable output variance bounds are also computed and a parametric study of design parameters is conducted to evaluate their influence on the OVC performance. The performance of the OVC control method is evaluated both on the linear and nonlinear OCT models. Furthermore, control design for the OCT with sensor failures is also examined. Lastly, the MPC strategy is also investigated to improve the OCT flight control performance in simultaneous satisfaction of multiple constraints and to avoid blade stall. / Ph. D.

We propose a Model-Based Deep Reinforcement Learning (MBDRL) framework for collaborative paylaod transportation using Unmanned Aerial Vehicles (UAVs) in Search and Rescue (SAR) missions, enabling heavier payload conveyance while maintaining vehicle agility. Our approach extends the single-drone application to a novel multi-drone one, using the Probabilistic Ensembles with Trajectory Sampling (PETS) algorithm to model the unknown stochastic system dynamics and uncertainty. We use the Multi-Agent Reinforcement Learning (MARL) framework via a centralized controller in a leader-follower configuration. The agents utilize the approximated transition function in a Model Predictive Controller (MPC) configured to maximize the reward function for waypoint navigation, while a position-based formation controller ensures stable flights of these physically linked UAVs. We also developed an Unreal Engine (UE) simulation connected to an offboard planner and controller via a Robot Operating System (ROS) framework that is transferable to real robots. This work achieves stable waypoint navigation in a stochastic environment with a sample efficiency following that seen in single UAV work. This work has been funded by the National Science Foundation (NSF) under Award No. 2046770. / Master of Science / We apply the Model-Based Deep Reinforcement Learning (MBDRL) framework to the novel application of a UAV team transporting a suspended payload during Search and Rescue missions. Collaborating UAVs can transport heavier payloads while staying agile, reducing the need for human involvement. We use the Probabilistic Ensemble with Trajectory Sampling (PETS) algorithm to model uncertainties and build on the previously used single UAVpayload system. By utilizing the Multi-Agent Reinforcement Learning (MARL) framework via a centralized controller, our UAVs learn to transport the payload to a desired position while maintaining stable flight through effective cooperation. We also develop a simulation in Unreal Engine (UE) connected to a controller using a Robot Operating System (ROS) architecture, which can be transferred to real robots. Our method achieves stable navigation in unpredictable environments while maintaining the sample efficiency observed in single UAV scenarios.