Interpreting natural language actions in a simulated world is the first step towards robots controlled by natural language commands. In this work we present several models for interpreting unrestricted natural language commands in a simple block world. We present and compare rule-based models and recurrent neural network models of various architectures. We also discuss strategies to deal with errors in natural language data and compare them. On the Language Grounding dataset, our models outperform the previous state-of-the-art results in both source and location prediction reaching source accuracy 98.8% and average distance 0.71 between the correct and predicted location.

This thesis is focused on realization of the swarm intelligence experiments in the simulation software Vortex and MRDS. The aim is to decide whether the achieved results meet the theoretical expectations based on the published experiment.

The goal of this thesis is to create an artificial agent for an autonomous racing vehicle. This project is inspired by the F1/10 racing competition. The agent uses a planning algorithm to find a time-optimal trajectory. To achieve real-time performance, the agent analyzes the map of the track and it plans only for the next two corners immediately ahead of the vehicle. The agent re-plans several times per second as it drives along the circuit to account for imprecise trajectory following. We successfully tested the agent in the Gazebo simulator with good results. We also tested the algorithm on a custom car-like robot equipped with an on-board computer and sensors, but with limited success. 1

<p>U tezi je prikazan novi metod za sintezu i realizaciju dvonožnog<br />veštačkog hoda koji se zasniva na upotrebi jednostavnih pokreta čijim<br />je kombinovanjem moguće realizovati kompleksne pokrete kao što je<br />hod, a čiji se parametri mogu menjati tokom kretanja. Time je omogućeno<br />da se na osnovu informacija o nameravanom kretanju i stanja okoline<br />izvrši sinteza kretanja izborom i kombinacijom jednostavnih<br />bazičnih pokreta koje se nazivaju primitivi. Takođe je omogućeno da se,<br />tokom izvršavanja hoda bez njegovog prekida, menjaju parametri<br />kretanja kao što su brzina hoda, dužina koraka, pravac kretanja i<br />visina podizanja noge tokom prenosne faze. Potvrda je data kroz<br />eksperimentalne rezultate koji su sprovedeni simulacijom na<br />dinamičkom modelu humanoidnog robota.</p> / <p>This dissertation presents new method for the synthesis and realization of<br />biped artificial walk based on the use of simple movements that can be<br />combined in order to achieve complex movements such as walk, whereas it<br />is possible to change the motion parameters at any time. It means that,<br />based on the information about intended movement and current state of the<br />environment, it is possible to synthesize motion by selecting and tying simple<br />movements, i.e. motion primitives. It also enables the robot to change<br />walking parameters online such as walking speed, direction of walk, foot<br />length during swing phase and step length. Proof of this method is given by<br />experimental results obtained during the simulation on a dynamic model of<br />humanoid robot.</p>

<p>U bežičnoj mreži senzora i robota, senzorski moduli vrše nadzor<br />fizičkih veličina od značaja, a roboti imaju ulogu izvršilaca<br />zadataka koji im se dodeljuju primenom odgovarajućeg algoritma. Nakon<br />detekcije događaja od strane statičkih senzorskih čvorova i<br />prosleđivanja informacija o događajima robotima, potrebno je<br />dodeliti zadatke robotima na efikasan način. Dodela zadataka vrši<br />se u skladu sa prirodom različitih scenarija koji se mogu javiti u<br />praksi. U okviru disertacije razmatran je slučaj kada se konkurentno<br />javlja više događaja kojima je potrebno dodeliti izvršioce. U pogledu<br />energetske efikasnosti, u ovakvim sistemima kao ključni problemi<br />javljaju se minimizacija ukupne dužine kretanja robota i optimizacija<br />komunikacije u mreži. Od komunikacinih protokola za otkrivanje<br />izvršilaca, u ovoj disertaciji predstavljena su poboljšanja<br />postojećeg iMesh protokola i uveden je novi vCell protokol zasnovan na<br />lokalizovanom formiranju ćelija Voronoi dijagrama. Takođe,<br />upoređene su performanse novog protokola sa postojećim (pravougaoni<br />kvorum i iMesh) u gustim mrežama, retkim mrežama i mrežama sa<br />rupama u topologiji. Uz to, uvedeni su algoritmi za ažuriranje lokacije<br />kojima mreža reaguje na kretanje robota. Rezultati simulacija pokazuju<br />da vCell postiže efikasnost blizu 100% u nalaženju najbližeg robota u<br />gustim mrežama. U retkim mrežama, efikasnost mu je do 40% bolja u<br />odnosu na ostala rešenja.</p><p>Kao glavni rezultat u disertaciji prikazani su novi algoritmi za<br />dodelu robota kao izvršilaca zadataka događajima, čime su<br />prevaziđni nedostaci više do sada poznatih rešenja ovog problema.<br />Za zadati skup događaja i skup robota, svakom događaju dodeljen je po<br />jedan robot koji je zadužen za obilazak lokacije događaja. Tokom<br />pojedinačnih rundi, robotima je dozvoljen obilazak jednog događaja<br />kada se vrši uparivanje, ili više događaja, kada se vrši<br />sekvencijalna dodela. U distribuiranom slučaju, statički senzorski<br />uređaji detektuju događaje i prijavljuju ih obližnjim robotima.<br />Algoritam PDM koji se odnosi na unapređeno uparivanje sa mogućnošću<br />razmene partnera, eliminiše dugačke ivice koje se mogu javiti<br />prilikom uparivanja. Algoritam SQD za sekvencijalnu dodelu događaja<br />robotima iterativno pronalazi par robot-događaj sa najmanjim<br />međusobnim rastojanjem, uvrštava izabrani događaj u listu za oblazak<br />izabranog robota i ažurira poziciju robota. Takođe su predložene<br />generalizacije koje omogućavaju da događaji budu posećeni od strane<br />više robota i koje uzimaju u obzir vremenska ograničenja.<br />Distribuirani algoritam MAD, koji je zasnovan na iMesh<br />informacionoj strukturi i lokalnim aukcijama u robotskoj mreži,<br />vrši dodelu robota događajima na lokalizovan i energetski efikasan<br />način. Rezultati simulacija potvrđuju prednosti predloženih<br />algoritama u odnosu na postojeća rešenja, kako u pogledu skraćivanja<br />dužina putanja robota, tako i u produženju životnog vremena sistema.</p> / <p>In a typical wireless sensor and robot network, sensor nodes monitor physical<br />values of interest, while robots perform some automated tasks. The tasks are<br />assigned to robots by means of an appropriate algorithm. Upon the<br />occurrence of events which are detected by sensor nodes, the information<br />about the events needs to be delivered to robots. Afterwards, it is necessary<br />to assign tasks to robots in an efficient way. Task assignment is performed<br />according to the nature of different scenarios which might occur in practice.<br />This thesis is focused on the case when multiple events, all of which require<br />to be visited by robots, happen simultaneously. Regarding energy efficiency,<br />the key issues which arise in such systems are minimization of robot travel<br />paths, and optimization of the network traffic. In this thesis, the following<br />service discovery protocols are presented: improvements of the existing<br />iMesh protocol, and the novel vCell protocol, which is based on localized<br />formation of an information structure which resembles Voronoi diagram.<br />Furthermore, the performaces of new vCell protocol is compared with the<br />existing protocols (Quorum and iMesh) in dense networks, sparse networks,<br />and networks with holes in topology. Also, location update algorithms are<br />introduced, which deal with robot mobility. The simulations show that vCell<br />achieves nearly 100% success rate in finding the nearest robot in dense<br />networks. In sparse networks, it outperforms the other existing solutions by up<br />to 40%.<br />As a key contributtion, the novel dispatch lgorithms have been introduced.<br />Given a set of events and a set of robots, the dispatch problem is to allocate<br />one robot for each event to visit it. In a single round, each robot may be<br />allowed to visit only one event (matching dispatch), or several events in a<br />sequence (sequence dispatch). In a distributed setting, each event is<br />discovered by a sensor and reported to a robot. In this thesis, novel<br />algorithms are presented, whichh are aimed at overcoming the shortcomings<br />of several existing solutions. Pairwise distance based matching algorithm<br />(PDM) eliminates long edges by pairwise exchanges between matching pairs.<br />Sequence dispatch algorithm (SQD) iteratively finds the closest event-robot<br />pair, includes the event in dispatch schedule of the selected robot and<br />updates its position accordingly. When event-robot distances are multiplied by<br />robot resistance (inverse of the remaining energy), the corresponding energybalanced<br />variants are obtained. Also, generalizations are introduced which<br />handle multiple visits and timing constraints. Distributed algorithm MAD is<br />based on information mesh infrastructure and local auctions within the robot<br />network for obtaining the optimal dispatch schedule for each robot. The<br />simulations conducted confirm the advantages of our algorithms over other<br />existing solutions in terms of average robot-event distance and lifetime.</p>

Multi-Agent Path Finding (MAPF) is the task to find efficient collision-free paths for a fixed set of agents. Each agent moves from its initial location to its desired destination in a shared environment represented by a graph. The classical definition of MAPF is very simple and usually does not reflect the real world accurately. In this thesis, we try to add several attributes to the MAPF definition so that we overcome this shortcoming. This is done in several steps. First, we present an approach on how to model and solve MAPF via reduction to Boolean satisfiability using Picat programming language. This provides us with a useful model that can be easily modified to accommodate additional constraints. Secondly, we modify MAPF to portray a more realistic world. Specifically, we allow new agents to enter the shared environment during the execution of the found plan, and we relax the requirement on the homogeneousness of the shared environment. Lastly, we experimentally verify the applicability of the novel models on real robots in comparison with the classical MAPF setting.

This diploma thesis deals with roboethics and its associated moral and social dilemmas. The theoretical part of this thesis defines roboethics and presents the diverse opinions and approaches, deals with the position of robots in society and also subsequently examines the various problematic areas of the use of robots. The conclusion of the theoretical part of the thesis then focuses on autonomous vehicles, to which the quantitative and qualitative research parts of the thesis are devoted. The research part of this diploma thesis aims to find out the opinions on this issue of the Czech general public and Czech experts as well as to find out whether their opinions are aligned, how their opinions differ and lastly, which issues are the most important for the future development of this field.

This thesis addresses the issue of autonomous weapons systems and their potential preventive prohibition with regard to current international discussions at multilateral forums such as the Human Rights Council, First Committee of the General Assembly, and Convention on Certain Conventional Weapons at UN. The aim of this thesis is to provide an extensive empirical account of the substance of those discussions and their most likely outcome, estimating state preferences with use of content analysis and the likely outcome with median voter prediction. From a theoretical standpoint, the thesis draws from defensive realism and contributions of arms control, arms trade as well as institutionalist literature from which it draws the concept of legalization. From a methodological standpoint, the thesis relies on quantitative methods, in particular, content analysis for collection of data and median voter theorem for prediction of the likely outcome. In addition, the thesis uses the method of regression analysis to examine states' activity at the aforementioned fora. In conclusion, the thesis finds that the most likely outcome of discussions on autonomous weapons systems is a moderate-obligation form of hybrid regulation, which includes solutions such as framework convention and moratorium. Further finding of...

Terraforming mars by autonomous robotic building system, inspired by social insect. Idea design of robotic solution, their transportation and software. Design of structures builded by robots and their function. Pyramidal structures interwoven with ice tunnel, heat the atmosphere of Mars and accumulate heat in their mass. These structures are built with cheap and simple robots that mimic the behavior of social insects.

Poslední roky přinesly rostoucí trend nasazení robotů v nových aplikacích, kde se od nich očekává nejen práce vedle lidí, ale skutečná spolupráce na společných komplexních úlohách. K umožnění blízké, bohaté a přirozené interakce člověka s robotem, bude nutné podstatně rozšířit schopnosti současných robotických systémů. Dále již nebude docházet k interakci jen mezi roboty v bezpečnostních klecích a experty na jejich programování. Stále častěji budou interagovat s~bezpečnými spolupracujícími roboty uživatelé bez odborných znalostí z oblasti robotiky, s různorodým vzděláním a zkušenostmi. Úžasně složitá zařízení, kterými dnešní roboti jsou, se tak stanou ještě složitějšími, což představuje zásadní výzvu pro návrh jejich uživatelských rozhraní. Cílem této práce je zkoumat a vyvinout řešení umožňující blízkou interakci mezi neodbornými uživateli a komplexními roboty. Výzkum byl zaměřen na dvě oblasti robotiky: asistenční servisní a průmyslové spolupracující roboty. Ačkoliv se tyto dvě oblasti vyznačují odlišnými požadavky, pro návrh interakce mezi člověkem a~robotem je možné použít podobné principy. Nedostatky stávajících přístupů jsou řešeny návrhem nové metody pro úlohově zaměřenou interakci. Nejvýznamější aspekty metody jsou využití smíšené reality, autonomních funkcí robota, komunikace vnitřního stavu robota, kontextová citlivost a použití modalit vhodných pro danou úlohu. Pro obě oblasti zaměření výzkumu bylo na základě metody navrženo a implementováno uživatelské rozhraní. Obě rozhraní byla úspěšně ověřena s neodbornými uživateli, kteří díky nim byli schopni úspěšně spolupracovat s robotem na složitých úlohách. Publikovaná ověření rozhraní prokazují, že realizovaná metoda významně zlepšuje blízkou interakci mezi člověkem a robotem, která s dosavadními přístupy nebyla plně dosažitelná. Klíčové aspekty metody představují vodítko pro návrh uživatelských rozhraní v oblasti spolupracujících robotů.