This thesis presents a framework for a mobile robot to escort a human to their destination successfully and efficiently. The proposed technique uses accurate path prediction incorporating human intention to locate the robot in front of the human while walking. Human intention is inferred by the head pose, an effective past-proven implicit indicator of intention, and fused with conventional physics-based motion prediction. The human trajectory is estimated and predicted using a particle filter because of the human's nonlinear and non-Gaussian behavior, and the robot control action is determined from the predicted human pose allowing for anticipative autonomous escorting. Experimental analysis shows that the incorporation of the proposed human intention model reduces human position prediction error by approximately 35% when turning. Furthermore, experimental validation with an omnidirectional mobile robotic platform shows escorting up to 50% more accurate compared to the conventional techniques, while achieving 97% success rate. / Master of Science / This thesis presents a method for a mobile robot to escort a human to their destination successfully and efficiently. The proposed technique uses human intention to predict the walk path allowing the robot to be in front of the human while walking. Human intention is inferred by the head direction, an effective past-proven indicator of intention, and is combined with conventional motion prediction. The robot motion is then determined from the predicted human position allowing for anticipative autonomous escorting. Experimental analysis shows that the incorporation of the proposed human intention reduces human position prediction error by approximately 35% when turning. Furthermore, experimental validation with an mobile robotic platform shows escorting up to 50% more accurate compared to the conventional techniques, while achieving 97% success rate. The unique escorting interaction method proposed has applications such as touch-less shopping cart robots, exercise companions, collaborative rescue robots, and sanitary transportation for hospitals.

No description available.

風險偏好與預測精確性對生存力的重要性吸引進來許多理論學者的注意。一個極端是認為風險偏好完全不重要，唯一重要是預測精確性。然而此乃基於柏拉圖最適配置之下。透過代理人基模型，我們發現相異的結果，即風險偏好在生存力上扮演重要角色。 / The relevance of risk preference and forecasting accuracy to the survival of investors is an issue that has recently attracted a number of recent theoretical studies. At one extreme, it has been shown that risk preference can be entirely irrelevant, and that in the long run what distinguishes the agents who survive from those who vanish is just their forecasting accuracy. Being in line with the market selection hypothesis, this theoretical result is, however, established mainly on the basis of Pareto optimal allocation. By using agent-based computational modeling, this dissertation extends the existing studies to an economy where adaptive behaviors are autonomous and complex heterogeneous, and where the economy is notorious for its likely persistent deviation from Pareto optimality. Specifically, a computational multiasset artificial stock market corresponding to Blume and Easley (1992) and Sandroni (2000) is constructed and studied. Through simulation, we present results that contradict the market selection hypothesis. Risk preference plays a key role in survivability. And agents who have superior forecasting accuracy may be driven out just because of their risk preference. Nevertheless, when all the agents are with the same preference, the wealth share is positively correlated to forecasting accuracy, and the market selection hypothesis is sustained, at least in a weak sense.

Perception-action (P-A) learning is an approach to cognitive system building that seeks to reduce the complexity associated with conventional environment-representation/action-planning approaches. Instead, actions are directly mapped onto the perceptual transitions that they bring about, eliminating the need for intermediate representation and significantly reducing training requirements. We here set out a very general learning framework for cognitive systems in which online learning of the P-A mapping may be conducted within a symbolic processing context, so that complex contextual reasoning can influence the P-A mapping. In utilizing a variational calculus approach to define a suitable objective function, the P-A mapping can be treated as an online learning problem via gradient descent using partial derivatives. Our central theoretical result is to demonstrate top-down modulation of low-level perceptual confidences via the Jacobian of the higher levels of a subsumptive P-A hierarchy. Thus, the separation of the Jacobian as a multiplying factor between levels within the objective function naturally enables the integration of abstract symbolic manipulation in the form of fuzzy deductive logic into the P-A mapping learning. We experimentally demonstrate that the resulting framework achieves significantly better accuracy than using P-A learning without top-down modulation. We also demonstrate that it permits novel forms of context-dependent multilevel P-A mapping, applying the mechanism in the context of an intelligent driver assistance system. / DIPLECS / GARNICS / CUAS

CoordenaÃÃo de AperfeÃoamento de Pessoal de NÃvel Superior / O senso de imersÃo do usuÃrio em um ambiente virtual requer nÃo somente alta qualidade visual grÃfica, mas tambÃm comportamentos adequados por parte dos personagens virtuais, isto Ã, com movimentos e aÃÃes que correspondam Ãs suas caracterÃsticas fÃsicas e aos eventos que ocorrem em seu meio. Nesse contexto, percebe-se o papel fundamental desempenhado pelo modo como os agentes se comportam em aplicaÃÃes de RV. O problema que permanece em aberto Ã: âComo obter comportamentos autÃnomos naturais e realistas de personagens virtuais?â. Um agente à dito autÃnomo se ele for capaz de gerar suas prÃprias normas (do grego autos, "a si mesmo", e nomos, "norma", "ordem"). Logo, autonomia implica em aÃÃes realizadas por um agente que resultam da estreita interaÃÃo entre suas dinÃmicas internas e os eventos ocorrendo no ambiente ao seu redor, ao invÃs de haver um controle externo ou uma especificaÃÃo de respostas em um plano prÃ-definido. Desse modo, um comportamento autÃnomo deveria refletir os detalhes da associaÃÃo entre o personagem e o ambiente, implicando em uma maior naturalidade e realismo nos movimentos. Assim, chega-se à proposta de que um comportamento à considerado natural se ele mantÃm coerÃncia entre o corpo do personagem e o ambiente ao seu redor. Para um observador externo, tal coerÃncia à percebida como comportamento inteligente. Essa noÃÃo resulta do atual debate, no campo da InteligÃncia Artificial, sobre o significado da inteligÃncia. Baseado nas novas tendÃncias surgidas dessas discussÃes, argumenta-se que o nÃvel de coerÃncia necessÃrio a um comportamento natural apenas pode ser alcanÃado atravÃs de tÃcnicas de emergÃncia. AlÃm da defesa conceitual da abordagem emergentista para a geraÃÃo de comportamento de personagens virtuais, este estudo apresenta novas tÃcnicas para a implementaÃÃo dessas ideias. Entre as contribuiÃÃes, està a proposta de um novo processo de codificaÃÃo e evoluÃÃo de Redes Neurais Artificiais que permite o desenvolvimento de controladores para explorar as possibilidades da geraÃÃo de comportamentos por emergÃncia. TambÃm à explorada a evoluÃÃo sem objetivo, atravÃs da simulaÃÃo da reproduÃÃo sexuada de personagens. Para validar a tese, foram desenvolvidos experimentos envolvendo um robà virtual. Os resultados apresentados mostram que a auto-organizaÃÃo de um sistema à de fato capaz de produzir um acoplamento Ãntimo entre agente e ambiente. Como consequÃncia da abordagem adotada, foram obtidos comportamentos bastante coerentes com as capacidades dos personagens e as condiÃÃes ambientais, com ou sem descriÃÃo de objetivos. Os mÃtodos propostos se mostraram sensÃveis a modificaÃÃes do ambiente e a modificaÃÃes no sensoriamento do robÃ, comprovando robustez ao gerar cÃrtices visuais funcionais, seja com sensores de proximidade, seja com cÃmeras virtuais, interpretando seus pixels. Ressalta-se tambÃm a geraÃÃo de diferentes tipos de comportamentos interessantes, sem qualquer descriÃÃo de objetivos, nos experimentos envolvendo reproduÃÃo simulada. / The userâs sense of immersion requires not only high visual quality of the virtual environment, but also accurate simulations of dynamics to ensure the reliability of the experience. In this context, the way the characters behave in a virtual environment plays a fundamental role. The problem that remains open is: âWhat needs to be done for autonomous virtual characters to display natural/realistic behaviors?â. A behavior is considered autonomous when the actions performed by the agent result from a close interaction between its internal dynamics and the circumstantial events in the environment, rather than from external control or specification dictated by a predefined plan. Thus, an autonomous behavior should reflect the details of the association between the character and its environment, resulting in greater naturalness and realistic movements. Therefore, it is proposed that the behavior is considered natural if it maintains coherence between the characterâs body and the environment surrounding it. To an external observer, such coherence is perceived as intelligent behavior. This notion of intelligent behavior arose from a current debate, in the field of Artificial Intelligence, about the meaning of intelligence. Based on the new trends that came out from those discussions, it is argued that the level of coherence required for natural behavior in complex situations can only be achieved through emergence. In addition to the conceptual support of the emergentist approach to generating behavior of virtual characters, this study presents new techniques for implementing those ideas. A contribution of this work is a novel technique for the enconding and evolution of Artificial Neural Networks, which allows the development of controllers to explore the possibilities of generating behaviors through emergence. Evolution without objective description is also explored through the simulation of sexual reproduction of characters. In order to validate the theory, experiments involving a virtual robot were developed. The results show that self-organization of a system is indeed able to produce an intimate coupling between agent and environment. As a consequence of the adopted approach, it were achieved behaviors quite consistent with the characterâs capabilities and environmental conditions, with or without description of objectives. The proposed methods were sensitive to changes in the environment and in the robotâs sensory apparatus, proving robustness on generating functional visual cortices, either with proximity sensors or with virtual cameras, interpreting its pixels. It is also emphasized the generation of different types of interesting behaviors, without any description of objectives, in experiments involving simulated reproduction.

Detta examensarbete undersöker navigering för grupper av autonoma agenter i dataspelsmiljöer. Genom att kombinera olika styrbeteenden och beräkningsmodeller utvärderar arbetet vilken av dessa tekniker som är mest effektiv med avseende på tid och vägval i trånga spelmiljöer. En experimentmiljö har utvecklats som implementerar fyra stycken tekniker och utvärderar dessa i tre olika miljöer med 10 respektive 50 agenter som navigerar genom miljön. Som grund använder samtliga tekniker ett vägföljningsbeteende och ett flockbeteende. Det som skiljer teknikerna åt är vilken beräkningsmodell som används samt att två av teknikerna använder ett väggundvikelsebeteende. Resultatet visar att alla tekniker är användbara men att den mer avancerade beräkningsmodellen ger ett bättre resultat överlag. Väggundvikelsebeteendet bidrar också till ett bättre resultat och gör alltså nytta i de miljöer som använts. Ett problem med styrbeteenden är dock balanseringen av vikterna som används i teknikerna och det kan krävas mycket finjustering innan man får ett bra beteende.

Behavior Trees (BTs) are a reactive policy representation that has gained popularity in recent years, especially in the robotics domain. Among the learning methods for BTs, Genetic Programming (GP) is an effective method for learning a good BT. One drawback of GP is that it is likely to get stuck in local minima. In this project, we focus on studying both the existing methods and new directions to avoid local minima and improve the efficiency of learning BT with GP. The methods studied in the project are the grid search, the Bayesian Optimization (BO), the Distributed Island Model (DIM) and the dynamic selection pressure. We performed the experiments with four different benchmark applications implemented with high-level state machines. The changes related to fitness values, diversity, and origin throughout the learning processes were collected and analyzed as part of the quantitative analysis. Some generated BTs were selected for the qualitative analysis to provide insights into the local minima and individuals with ideal performance. Based on our experiments, we conclude that learning BTs with GP can benefit from a fitness function that is sensitive to the performance differences of the individuals. The effect of methods including the DIM and the dynamic selection pressure depends on both the applications and the settings. We recommend the grid search method for hyperparameter searching and the DIM for accelerating the learning process from distributed computing. / BTs är en reaktiv policy-representation som har ökat i popularitet de senaste åren, särskilt inom robotik. Bland inlärningsmetoderna för BTs är GP en effektiv metod för att generera bra BT. En nackdel med GP är att den lätt fastnar i lokala minima. I det här projektet fokuserar vi på att studera på existerande metoder och nya sätt att undvika lokala minima och öka inlärningseffektiviteten för BT med GP. Metoderna som studerats i projektet är grid search, BO, DIM och dynamic selection pressure. Vi genomförde experiment med fyra olika benchmarkapplikationer som implementerats med högnivå-tillståndsmaskiner. Ändringar i fitnessvärden, mångfald och källa till ändringen genom inlärningsprocessen samlades in och analyserades genom kvantitativ analys. Några genererade BTs valdes ut för kvalitativ analys för att ge insikter i de lokala minimumen och vilka individer som ger ideal prestanda. Baserat på våra experiment konkluderar vi att inlärning av BTs med GP kan tjäna på en bra fitnessfunktion som är känslig för prestandaskillnader mellan invidider. Effekten av metoderna DIM och dynamic selection pressure beror på applikationen och inställningarna. Vi rekommenderar grid search för hyperparametersökning och DIM för att accelerera inlärningen från distribuerade system.

Simulation is an important tool for prediction and assessment of the behavior of complex systems and situations. The importance of simulation has increased tremendously during the last few decades, mainly because the rapid pace of development in the field of electronics has turned the computer from a costly and obscure piece of equipment to a cheap ubiquitous tool which is now an integral part of our daily lives. While such technological improvements make it easier to analyze well-understood deterministic systems, increase in speed and storage capacity alone are not enough when simulating situations where human beings and their behavior are an integral part of the system being studied. The problem with simulation of intelligent entities is that intelligence is still not well understood and it seems that the field of Artificial Intelligence (AI) has a long way to go before we get computers to think like humans. Behavior-based agent modeling has been proposed in mid-80's as one of the alternatives to the classical AI approach. While used mainly for the control of specialized robotic vehicles with very specific sensory capabilities and limited intelligence, we believe that a behavior-based approach to modeling generic autonomous agents in complex environments can provide promising results. To this end, we are investigating a behavior-based model for controlling groups of collaborating and competing agents in a geographic terrain. In this thesis, we are focusing on scenarios of military nature, where agents can move within the environment and adversaries can eliminate each other through use of weapons. Different aspects of agent behavior like navigation to a goal or staying in group formation, are implemented by distinct behavior modules and the final observed behavior for each agent is an emergent property of the combination of simple behaviors and their interaction with the environment. Our experiments show that while such an approach is quite efficient in terms of computational power, it has some major drawbacks. One of the problems is that reactive behavior-based navigation algorithms are not well suited for environments with complex mobility constraints where they tend to perform much worse than proper path planning. This problem represents an important research question, especially when it is considered that most of the modern military conflicts and operations occur in urban environments. One of the contributions of this thesis is a novel approach to reactive navigation where goals and terrain information are fused based on the idea of transforming a terrain with obstacles into a virtual obstacle-free terrain. Experimental results show that our approach can successfully combine the low run-time computational complexity of reactive methods with the high success rates of classical path planning. Another interesting research problem is how to deal with the unpredictable nature of emergent behavior. It is not uncommon to have situations where an outcome diverges significantly from the intended behavior of the agents due to highly complex nonlinear interactions with other agents or the environment itself. Chances of devising a formal way to predict and avoid such abnormalities are slim at best, mostly because such complex systems tend to be be chaotic in nature. Instead, we focus on detection of deviations through tracking group behavior which is a key component of the total situation awareness capability required by modern technology-oriented and network-centric warfare. We have designed a simple and efficient clustering algorithm for tracking of groups of agent suitable for both spatial and behavioral domain. We also show how to detect certain events of interest based on a temporal analysis of the evolution of discovered clusters.

<p dir="ltr">The growth of electric vehicles (EVs) offers several benefits for air quality improvement and emissions reduction. Nonetheless, EVs also pose several challenges in the area of highway transportation. These barriers are related to the limitations of EV technology, particularly the charge duration and speed of battery recharging, which translate to vehicle range anxiety for EV users. A promising solution to these concerns is V2V DWC technology (Vehicle to Vehicle Dynamic Wireless Charging), particularly mobile energy disseminators (MEDs). The MED is mounted on a large vehicle or truck that charges all participating EVs within a specified locus from the MED. However, current research on MEDs offers solutions that are widely considered impractical for deployment, particularly in urban environments where range anxiety is common. Acknowledging such gap in the literature, this thesis proposes a comprehensive methodological framework for optimal MED deployment decisions. In the first component of the framework, a practical system, termed “ChargingEnv” is developed using reinforcement learning (RL). ChargingEnv simulates the highway environment, which consists of streams of EVs and an MED. The simulation accounts for a possible misalignment of the charging panel and incorporates a realistic EV battery model. The second component of the framework uses multiple deep RL benchmark models that are trained in “ChargingEnv” to maximize EV service quality within limited charging resource constraints. In this study, numerical experiments were conducted to demonstrate the MED deployment decision framework’s efficacy. The findings indicate that the framework’s trained model can substantially improve EV travel range and alleviate battery depletion concerns. This could serve as a vital tool that allows public-sector road agencies or private-sector commercial entities to efficiently orchestrate MED deployments to maximize service cost-effectiveness.</p>

<p dir="ltr">The concept of reinforcement learning has become increasingly relevant in learning- based applications, especially in the field of autonomous navigation, because of its fundamental nature to operate without the necessity of labeled data. However, the infeasibility of training reinforcement learning based autonomous navigation applications in a real-world setting has increased the popularity of researching and developing on autonomous navigation systems by creating simulated environments in game engine platforms. This thesis investigates the comparative performance of Unity and Unreal Engine 5 within the framework of a reinforcement learning system applied to autonomous race car navigation. A rudimentary simulated setting featuring a model car navigating a racetrack is developed, ensuring uniformity in environmental aspects across both Unity and Unreal Engine 5. The research employs reinforcement learning with genetic algorithms to instruct the model car in race track navigation; while the tools and programming methods for implementing reinforcement learning vary between the platforms, the fundamental concept of reinforcement learning via genetic algorithms remains consistent to facilitate meaningful comparisons. The implementation includes logging of key performance variables during run times on each platform. A comparative analysis of the performance data collected demonstrates Unreal Engine's superior performance across the collected variables. These findings contribute insights to the field of autonomous navigation systems development and reinforce the significance of choosing an optimal underlying simulation platform for reinforcement learning applications.</p>