1. |
- Andersson, Olov, 1979-
(författare)
-
Learning to Make Safe Real-Time Decisions Under Uncertainty for Autonomous Robots
- 2020
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Robots are increasingly expected to go beyond controlled environments in laboratories and factories, to act autonomously in real-world workplaces and public spaces. Autonomous robots navigating the real world have to contend with a great deal of uncertainty, which poses additional challenges. Uncertainty in the real world accrues from several sources. Some of it may originate from imperfect internal models of reality. Other uncertainty is inherent, a direct side effect of partial observability induced by sensor limitations and occlusions. Regardless of the source, the resulting decision problem is unfortunately computationally intractable under uncertainty. This poses a great challenge as the real world is also dynamic. It will not pause while the robot computes a solution. Autonomous robots navigating among people, for example in traffic, need to be able to make split-second decisions. Uncertainty is therefore often neglected in practice, with potentially catastrophic consequences when something unexpected happens. The aim of this thesis is to leverage recent advances in machine learning to compute safe real-time approximations to decision-making under uncertainty for real-world robots. We explore a range of methods, from probabilistic to deep learning, as well as different combinations with optimization-based methods from robotics, planning and control. Driven by applications in robot navigation, and grounded in experiments with real autonomous quadcopters, we address several parts of this problem. From reducing uncertainty by learning better models, to directly approximating the decision problem itself, all the while attempting to satisfy both the safety and real-time requirements of real-world autonomy.
|
|
2. |
- Präntare, Fredrik, 1990-
(författare)
-
Dividing the Indivisible : Algorithms, Empirical Advances, and Complexity Results for Value-Maximizing Combinatorial Assignment Problems
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Allocating resources, goods, agents (e.g., humans), expertise, production, and assets is one of the most influential and enduring cornerstone challenges at the intersection of artificial intelligence, operations research, politics, and economics. At its core—as highlighted by a number of seminal works [181, 164, 125, 32, 128, 159, 109, 209, 129, 131]—is a timeless question: How can we best allocate indivisible entities—such as objects, items, commodities, jobs, or personnel—so that the outcome is as valuable as possible, be it in terms of expected utility, fairness, or overall societal welfare? This thesis confronts this inquiry from multiple algorithmic viewpoints, focusing on the value-maximizing combinatorial assignment problem: the optimization challenge of partitioning a set of indivisibles among alternatives to maximize a given notion of value. To exemplify, consider a scenario where an international aid organization is responsible for distributing medical resources, such as ventilators and vaccines, and allocating medical personnel, including doctors and nurses, to hospitals during a global health crisis. These resources and personnel—inherently indivisible and non-fragmentable—necessitate an allocation process designed to optimize utility and fairness. Rather than using manual interventions and ad-hoc methods, which often lack precision and scalability, a rigorously developed and demonstrably performant approach can often be more desirable. With this type of challenge in mind, our thesis begins through the lens of computational complexity theory, commencing with an initial insight: In general, under prevailing complexity-theoretic assumptions (P ≠ NP), it is impossible to develop an efficient method guaranteeing a value-maximizing allocation that is better than “arbitrarily bad”, even under severely constraining limitations and simplifications. This inapproximability result not only underscores the problem’s complexity but also sets the stage for our ensuing work, wherein we develop novel algorithms and concise representations for utilitarian, egalitarian, and Nash welfare maximization problems, aimed at maximizing average, equitable, and balanced utility, respectively. For example, we introduce the synergy hypergraph—a hypergraph-based characterization of utilitarian combinatorial assignment—which allows us to prove several new state-of-the-art complexity results to help us better understand how hard the problem is. We then provide efficient approximation algorithms and (non-trivial) exponential-time algorithms for many hard cases. In addition, we explore complexity bounds for generalizations with interdependent effects between allocations, known as externalities in economics. Natural applications in team formation, resource allocation, and combinatorial auctions are also discussed; and a novel “bootstrapped” dynamic-programming method is introduced. We then transition from theory to practice as we shift our focus to the utilitarian variant of the problem—an incarnation of the problem particularly applicable to many real-world scenarios. For this variation, we achieve substantial empirical algorithmic improvements over existing methods, including industry-grade solvers. This work culminates in the development of a new hybrid algorithm that combines dynamic programming with branch-and-bound techniques that is demonstrably faster than all competing methods in finding both optimal and near-optimal allocations across a wide range of experiments. For example, it solves one of our most challenging problem sets in just 0.25% of the time required by the previous best methods, representing an improvement of approximately 2.6 orders of magnitude in processing speed. Additionally, we successfully integrate and commercialize our algorithm into Europa Universalis IV—one of the world’s most popular strategy games, with a player base exceeding millions. In this dynamic and challenging setting, our algorithm efficiently manages complex strategic agent interactions, highlighting its potential to improve computational efficiency and decision-making in real-time, multi-agent scenarios. This also represents one of the first instances where a combinatorial assignment algorithm has been applied in a commercial context. We then introduce and evaluate several highly efficient heuristic algorithms. These algorithms—while lacking provable quality guarantees—employ general-purpose heuristic and random-sampling techniques to significantly outperform existing methods in both speed and quality in large-input scenarios. For instance, in one of our most challenging problem sets, involving a thousand indivisibles, our best algorithm generates outcomes that are 99.5% of the expected optimal in just seconds. This performance is particularly noteworthy when compared to state-of-the-art industry-grade solvers, which struggle to produce any outcomes under similar conditions. Further advancing our work, we employ novel machine learning techniques to generate new heuristics that outperform the best hand-crafted ones. This approach not only showcases the potential of machine learning in combinatorial optimization but also sets a new standard for combinatorial assignment heuristics to be used in real-world scenarios demanding rapid, high-quality decisions, such as in logistics, real-time tactics, and finance. In summary, this thesis bridges many gaps between the theoretical and practical aspects of combinatorial assignment problems such as those found in coalition formation, combinatorial auctions, welfare-maximizing resource allocation, and assignment problems. It deepens the understanding of the computational complexities involved and provides effective and improved solutions for longstanding real-world challenges across various sectors—providing new algorithms applicable in fields ranging from artificial intelligence to logistics, finance, and digital entertainment, while simultaneously paving the way for future work in computational problem-solving and optimization.
|
|
3. |
- Tiger, Mattias, 1989-
(författare)
-
Safety-Aware Autonomous Systems : Preparing Robots for Life in the Real World
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Real‐world autonomous systems are expected to be increasingly deployed and operating in real‐world environments over the coming decades. Autonomous systems such as AI‐enabled robotic systems and intelligent transportation systems, will alleviate mundane human work, provide new services, and facilitate a smarter and more flexible infrastructure. The real‐ world environments affected include workplaces, public spaces, and homes. To ensure safe operations, in for example the vicinity of people, it is paramount that the autonomous systems are explainable, behave predictable, and can handle that the real world is ever changing and only partially observable. To deal with a dynamic and changing environment, consistently and safely, it is necessary to have sound uncertainty management. Explicit uncertainty quantification is fundamental to providing probabilistic safety guarantees that can also be monitored during runtime to ensure safety in new situations. It is further necessary for well‐grounded prediction and classification uncertainty, for achieving task effectiveness with high robustness and for dealing with unknown unknowns, such as world model divergence, using anomaly detection. This dissertation focuses on the notion of motion in terms of trajectories, from recognizing – to anticipating – to generating – to monitoring that it fulfills expectations such as predictability or other safety constraints during runtime. Efficiency, effectiveness, and safety are competing qualities, and in safety-critical applications the required degree of safety makes it very challenging to reach useful levels of efficiency and effectiveness. To this end, a holistic perspective on agent motion in complex and dynamic environments is investigated. This work leverage synergies in well‐founded formalized interactions and integration between learning, reasoning, and interaction, and demonstrate jointly efficient, effective, and safe capabilities for autonomous systems in safety‐critical situations.
|
|
4. |
- Andersson, Olov, 1979-
(författare)
-
Methods for Scalable and Safe Robot Learning
- 2017
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Robots are increasingly expected to go beyond controlled environments in laboratories and factories, to enter real-world public spaces and homes. However, robot behavior is still usually engineered for narrowly defined scenarios. To manually encode robot behavior that works within complex real world environments, such as busy work places or cluttered homes, can be a daunting task. In addition, such robots may require a high degree of autonomy to be practical, which imposes stringent requirements on safety and robustness. \setlength{\parindent}{2em}\setlength{\parskip}{0em}The aim of this thesis is to examine methods for automatically learning safe robot behavior, lowering the costs of synthesizing behavior for complex real-world situations. To avoid task-specific assumptions, we approach this from a data-driven machine learning perspective. The strength of machine learning is its generality, given sufficient data it can learn to approximate any task. However, being embodied agents in the real-world, robots pose a number of difficulties for machine learning. These include real-time requirements with limited computational resources, the cost and effort of operating and collecting data with real robots, as well as safety issues for both the robot and human bystanders.While machine learning is general by nature, overcoming the difficulties with real-world robots outlined above remains a challenge. In this thesis we look for a middle ground on robot learning, leveraging the strengths of both data-driven machine learning, as well as engineering techniques from robotics and control. This includes combing data-driven world models with fast techniques for planning motions under safety constraints, using machine learning to generalize such techniques to problems with high uncertainty, as well as using machine learning to find computationally efficient approximations for use on small embedded systems.We demonstrate such behavior synthesis techniques with real robots, solving a class of difficult dynamic collision avoidance problems under uncertainty, such as induced by the presence of humans without prior coordination. Initially using online planning offloaded to a desktop CPU, and ultimately as a deep neural network policy embedded on board a 7 quadcopter.
|
|
5. |
- Conte, Gianpaolo, 1974-
(författare)
-
Navigation Functionalities for an Autonomous UAV Helicopter
- 2007
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- This thesis was written during the WITAS UAV Project where one of the goals has been the development of a software/hardware architecture for an unmanned autonomous helicopter, in addition to autonomous functionalities required for complex mission scenarios. The algorithms developed here have been tested on an unmanned helicopter platform developed by Yamaha Motor Company called the RMAX. The character of the thesis is primarily experimental and it should be viewed as developing navigational functionality to support autonomous flight during complex real world mission scenarios. This task is multidisciplinary since it requires competence in aeronautics, computer science and electronics. The focus of the thesis has been on the development of a control method to enable the helicopter to follow 3D paths. Additionally, a helicopter simulation tool has been developed in order to test the control system before flight-tests. The thesis also presents an implementation and experimental evaluation of a sensor fusion technique based on a Kalman filter applied to a vision based autonomous landing problem. Extensive experimental flight-test results are presented.
|
|
6. |
- Conte, Gianpaolo, 1974-
(författare)
-
Vision-Based Localization and Guidance for Unmanned Aerial Vehicles
- 2009
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The thesis has been developed as part of the requirements for a PhD degree at the Artificial Intelligence and Integrated Computer System division (AIICS) in the Department of Computer and Information Sciences at Linköping University.The work focuses on issues related to Unmanned Aerial Vehicle (UAV) navigation, in particular in the areas of guidance and vision-based autonomous flight in situations of short and long term GPS outage.The thesis is divided into two parts. The first part presents a helicopter simulator and a path following control mode developed and implemented on an experimental helicopter platform. The second part presents an approach to the problem of vision-based state estimation for autonomous aerial platforms which makes use of geo-referenced images for localization purposes. The problem of vision-based landing is also addressed with emphasis on fusion between inertial sensors and video camera using an artificial landing pad as reference pattern. In the last chapter, a solution to a vision-based ground object geo-location problem using a fixed-wing micro aerial vehicle platform is presented.The helicopter guidance and vision-based navigation methods developed in the thesis have been implemented and tested in real flight-tests using a Yamaha Rmax helicopter. Extensive experimental flight-test results are presented.
|
|
7. |
- de Leng, Daniel, 1988-
(författare)
-
Robust Stream Reasoning Under Uncertainty
- 2019
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Vast amounts of data are continually being generated by a wide variety of data producers. This data ranges from quantitative sensor observations produced by robot systems to complex unstructured human-generated texts on social media. With data being so abundant, the ability to make sense of these streams of data through reasoning is of great importance. Reasoning over streams is particularly relevant for autonomous robotic systems that operate in physical environments. They commonly observe this environment through incremental observations, gradually refining information about their surroundings. This makes robust management of streaming data and their refinement an important problem.Many contemporary approaches to stream reasoning focus on the issue of querying data streams in order to generate higher-level information by relying on well-known database approaches. Other approaches apply logic-based reasoning techniques, which rarely consider the provenance of their symbolic interpretations. In this work, we integrate techniques for logic-based stream reasoning with the adaptive generation of the state streams needed to do the reasoning over. This combination deals with both the challenge of reasoning over uncertain streaming data and the problem of robustly managing streaming data and their refinement.The main contributions of this work are (1) a logic-based temporal reasoning technique based on path checking under uncertainty that combines temporal reasoning with qualitative spatial reasoning; (2) an adaptive reconfiguration procedure for generating and maintaining a data stream required to perform spatio-temporal stream reasoning over; and (3) integration of these two techniques into a stream reasoning framework. The proposed spatio-temporal stream reasoning technique is able to reason with intertemporal spatial relations by leveraging landmarks. Adaptive state stream generation allows the framework to adapt to situations in which the set of available streaming resources changes. Management of streaming resources is formalised in the DyKnow model, which introduces a configuration life-cycle to adaptively generate state streams. The DyKnow-ROS stream reasoning framework is a concrete realisation of this model that extends the Robot Operating System (ROS). DyKnow-ROS has been deployed on the SoftBank Robotics NAO platform to demonstrate the system's capabilities in a case study on run-time adaptive reconfiguration. The results show that the proposed system - by combining reasoning over and reasoning about streams - can robustly perform stream reasoning, even when the availability of streaming resources changes.
|
|
8. |
- de Leng, Daniel, 1988-
(författare)
-
Spatio-Temporal Stream Reasoning with Adaptive State Stream Generation
- 2017
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- A lot of today's data is generated incrementally over time by a large variety of producers. This data ranges from quantitative sensor observations produced by robot systems to complex unstructured human-generated texts on social media. With data being so abundant, making sense of these streams of data through reasoning is challenging. Reasoning over streams is particularly relevant for autonomous robotic systems that operate in a physical environment. They commonly observe this environment through incremental observations, gradually refining information about their surroundings. This makes robust management of streaming data and its refinement an important problem.Many contemporary approaches to stream reasoning focus on the issue of querying data streams in order to generate higher-level information by relying on well-known database approaches. Other approaches apply logic-based reasoning techniques, which rarely consider the provenance of their symbolic interpretations. In this thesis, we integrate techniques for logic-based spatio-temporal stream reasoning with the adaptive generation of the state streams needed to do the reasoning over. This combination deals with both the challenge of reasoning over streaming data and the problem of robustly managing streaming data and its refinement.The main contributions of this thesis are (1) a logic-based spatio-temporal reasoning technique that combines temporal reasoning with qualitative spatial reasoning; (2) an adaptive reconfiguration procedure for generating and maintaining a data stream required to perform spatio-temporal stream reasoning over; and (3) integration of these two techniques into a stream reasoning framework. The proposed spatio-temporal stream reasoning technique is able to reason with intertemporal spatial relations by leveraging landmarks. Adaptive state stream generation allows the framework to adapt in situations in which the set of available streaming resources changes. Management of streaming resources is formalised in the DyKnow model, which introduces a configuration life-cycle to adaptively generate state streams. The DyKnow-ROS stream reasoning framework is a concrete realisation of this model that extends the Robot Operating System (ROS). DyKnow-ROS has been deployed on the SoftBank Robotics NAO platform to demonstrate the system's capabilities in the context of a case study on run-time adaptive reconfiguration. The results show that the proposed system – by combining reasoning over and reasoning about streams – can robustly perform spatio-temporal stream reasoning, even when the availability of streaming resources changes.
|
|
9. |
- Heintz, Fredrik, 1975-
(författare)
-
DyKnow : A Stream-Based Knowledge Processing Middleware Framework
- 2009
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- As robotic systems become more and more advanced the need to integrate existing deliberative functionalities such as chronicle recognition, motion planning, task planning, and execution monitoring increases. To integrate such functionalities into a coherent system it is necessary to reconcile the different formalisms used by the functionalities to represent information and knowledge about the world. To construct and integrate these representations and maintain a correlation between them and the environment it is necessary to extract information and knowledge from data collected by sensors. However, deliberative functionalities tend to assume symbolic and crisp knowledge about the current state of the world while the information extracted from sensors often is noisy and incomplete quantitative data on a much lower level of abstraction. There is a wide gap between the information about the world normally acquired through sensing and the information that is assumed to be available for reasoning about the world.As physical autonomous systems grow in scope and complexity, bridging the gap in an ad-hoc manner becomes impractical and inefficient. Instead a principled and systematic approach to closing the sensereasoning gap is needed. At the same time, a systematic solution has to be sufficiently flexible to accommodate a wide range of components with highly varying demands. We therefore introduce the concept of knowledge processing middleware for a principled and systematic software framework for bridging the gap between sensing and reasoning in a physical agent. A set of requirements that all such middleware should satisfy is also described.A stream-based knowledge processing middleware framework called DyKnow is then presented. Due to the need for incremental refinement of information at different levels of abstraction, computations and processes within the stream-based knowledge processing framework are modeled as active and sustained knowledge processes working on and producing streams. DyKnow supports the generation of partial and context dependent stream-based representations of past, current, and potential future states at many levels of abstraction in a timely manner.To show the versatility and utility of DyKnow two symbolic reasoning engines are integrated into Dy-Know. The first reasoning engine is a metric temporal logical progression engine. Its integration is made possible by extending DyKnow with a state generation mechanism to generate state sequences over which temporal logical formulas can be progressed. The second reasoning engine is a chronicle recognition engine for recognizing complex events such as traffic situations. The integration is facilitated by extending DyKnow with support for anchoring symbolic object identifiers to sensor data in order to collect information about physical objects using the available sensors. By integrating these reasoning engines into DyKnow, they can be used by any knowledge processing application. Each integration therefore extends the capability of DyKnow and increases its applicability.To show that DyKnow also has a potential for multi-agent knowledge processing, an extension is presented which allows agents to federate parts of their local DyKnow instances to share information and knowledge.Finally, it is shown how DyKnow provides support for the functionalities on the different levels in the JDL Data Fusion Model, which is the de facto standard functional model for fusion applications. The focus is not on individual fusion techniques, but rather on an infrastructure that permits the use of many different fusion techniques in a unified framework.The main conclusion of this thesis is that the DyKnow knowledge processing middleware framework provides appropriate support for bridging the sense-reasoning gap in a physical agent. This conclusion is drawn from the fact that DyKnow has successfully been used to integrate different reasoning engines into complex unmanned aerial vehicle (UAV) applications and that it satisfies all the stated requirements for knowledge processing middleware to a significant degree.
|
|
10. |
- Landén, David, 1978-
(författare)
-
Complex Task Allocation for Delegation : From Theory to Practice
- 2011
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The problem of determining who should do what given a set of tasks and a set of agents is called the task allocation problem. The problem occurs in many multi-agent system applications where a workload of tasks should be shared by a number of agents. In our case, the task allocation problem occurs as an integral part of a larger problem of determining if a task can be delegated from one agent to another.Delegation is the act of handing over the responsibility for something to someone. Previously, a theory for delegation including a delegation speech act has been specified. The speech act specifies the preconditions that must be fulfilled before the delegation can be carried out, and the postconditions that will be true afterward. To actually use the speech act in a multi-agent system, there must be a practical way of determining if the preconditions are true. This can be done by a process that includes solving a complex task allocation problem by the agents involved in the delegation.In this thesis a constraint-based task specification formalism, a complex task allocation algorithm for allocating tasks to unmanned aerial vehicles and a generic collaborative system shell for robotic systems are developed. The three components are used as the basis for a collaborative unmanned aircraft system that uses delegation for distributing and coordinating the agents' execution of complex tasks.
|
|