Molecular structure generation is a fundamental problem that involves determining the 3D positions of molecules’ constituents. It has crucial biological applications, such as molecular docking, protein folding, and molecular design. Recent advances in generative modeling, such as diffusion models and flow matching, have made great progress on these tasks by modeling molecular conformations as a distribution. In this work, we focus on flow matching and adopt an energybased perspective to improve training and inference of structure generation models. Our view results in a mapping function, represented by a deep network, that is directly learned to iteratively map random configurations, i.e. samples from the source distribution, to target structures, i.e. points in the data manifold. This yields a conceptually simple and empirically effective flow matching setup that is theoretically justified and has interesting connections to fundamental properties such as idempotency and stability, as well as the empirically useful techniques such as structure refinement in AlphaFold. Experiments on protein docking as well as protein backbone generation consistently demonstrate the method’s effectiveness, where it outperforms recent baselines of task-associated flow matching and diffusion models, using a similar computational budget.

The electrification of society is an essential component of the effort to achieve a fossil-free world, where solar cells will play a central role in the future energy system. The advancement of photovoltaics should be aligned with principles of the circular economy. In the last years, lead halide perovskites have risen as a leading candidate for third-generation solar cells, experiencing rapid advancement. However, the manufacturing of commercial products inevitably generates significant waste and end-of-life devices, leading to potentially severe environmental repercussions. To tackle this challenge, proactive research and development of recycling and recovery technologies for perovskite solar cells are very necessary. Here we introduce a proof of concept, which, to the best of our knowledge, is the first AI-guided method designed to predict the optimal recycling treatment for perovskite solar cells and the first construction of a perovskite recycling dataset. Using sentiment analysis language processing for the first time, we achieved up to 70% accuracy in predicting the correct action for a given device structure. This innovative approach opens up new possibilities for applying sentiment analysis as a tool for e-waste recycling.

This paper outlines challenges and opportunities in operating underwater robots (so-called AUVs) on a seaweed farm. The need is driven by an emerging aquaculture industry on the Swedish west coast where large-scale seaweed farms are being developed. In this paper, the operational challenges are described and key technologies in using autonomous systems as a core part of the operation are developed and demonstrated. The paper presents a system and methods for operating an AUV in the seaweed farm, including initial localization of the farm based on a prior estimate and dead-reckoning navigation, and the subsequent scanning of the entire farm. Critical data from sidescan sonars for algorithm development are collected from real environments at a test site in the ocean, and the results are demonstrated in a simulated seaweed farm setup.

    In this paper, we show how Behavior Trees that have performance guarantees, in terms of safety and goal convergence, can be extended with components that were designed using machine learning, without destroying those performance guarantees.

    Machine learning approaches such as reinforcement learning or learning from demonstration can be very appealing to AI designers that want efficient and realistic behaviors in their agents. However, those algorithms seldom provide guarantees for solving the given task in all different situations while keeping the agent safe. Instead, such guarantees are often easier to find for manually designed model-based approaches. In this paper we exploit the modularity of behavior trees to extend a given design with an efficient, but possibly unreliable, machine learning component in a way that preserves the guarantees.    The approach is illustrated with an inverted pendulum example.

In this paper, we show how Behavior Trees that have performance guarantees, in terms of safety and goal convergence, can be extended with components that were designed using machine learning, without destroying those performance guarantees. Machine learning approaches such as reinforcement learning or learning from demonstration can be very appealing to AI designers that want efficient and realistic behaviors in their agents. However, those algorithms seldom provide guarantees for solving the given task in all different situations while keeping the agent safe. Instead, such guarantees are often easier to find for manually designed model-based approaches. In this paper we exploit the modularity of behavior trees to extend a given design with an efficient, but possibly unreliable, machine learning component in a way that preserves the guarantees. The approach is illustrated with an inverted pendulum example.

In this article, we provide a control-theoretic perspective on the research area of behavior trees in robotics. The key idea underlying behavior trees is to make use of modularity, hierarchies, and feedback in order to handle the complexity of a versatile robot control system. Modularity is a well-known tool to handle software complexity by enabling the development, debugging, and extension of separate modules without detailed knowledge of the entire system. A hierarchy of such modules is natural, since robot tasks can often be decomposed into a hierarchy of subtasks. Finally, feedback control is a fundamental tool for handling uncertainties and disturbances in any low-level control system, but in order to enable feedback control on the higher level, where one module decides what submodule to execute, information regarding the progress and applicability of each submodule needs to be shared in the module interfaces. We describe how these three concepts can be used in theoretical analysis, practical design, and extensions and combinations with other ideas from control theory and robotics.

Behavior trees represent a hierarchical and modular way of combining several low-level control policies into a high-level task-switching policy. Hybrid dynamical systems can also be seen in terms of task switching between different policies, and therefore several comparisons between behavior trees and hybrid dynamical systems have been made, but only informally, and only in discrete time. A formal continuous-time formulation of behavior trees has been lacking. Additionally, convergence analyses of specific classes of behavior tree designs have been made, but not for general designs. In this letter, we provide the first continuous-time formulation of behavior trees, show that they can be seen as discontinuous dynamical systems (a subclass of hybrid dynamical systems), which enables the application of existence and uniqueness results to behavior trees, and finally, provide sufficient conditions under which such systems will converge to a desired region of the state space for general designs. With these results, a large body of results on continuous-time dynamical systems can be brought to use when designing behavior tree controllers.

Critical robotic systems are systems whose functioning is critical to both ensuring the accomplishment of a given mission and preventing the endangerment of life and the surrounding environment. These critical aspects can be formally captured by convergence, in the sense that the system's state goes to a desired region of the statespace, and safety, in the sense that the system's state avoids unsafe regions of the statespace. Data-driven control policies, found through e.g. imitation learning or reinforcement learning, can outperform model-based methods in achieving convergence and safety efficiently; however, they often only do so by encouraging them, thus, they can be difficult to trust. Model-based control policies, on the other hand, are often well-suited to admitting formal guarantees of convergence and safety, thus they are often easier to trust. The main question asked in this thesis is: how can we compose data-driven and model-based control policies together to encourage efficiency while, at the same time, formally guaranteeing convergence and safety?

We answer this question with behaviour trees, a framework to represent hybrid control systems in a modular way. We present the first formal definition of behaviour trees as a hybrid system and present the conditions under which the execution of any behaviour tree as a hybrid control system will formally guarantee convergence and safety. Moreover, we present the conditions under which such formal guarantees can be maintained when including unguaranteed data-driven control policies, such as those coming from imitation learning or reinforcement learning. We also present an approach to synthesise such data-driven control policies in such a way that they encourage convergence and safety by adapting to unforeseen events. Alongside the above, we also explore an ancillary aspect of robot autonomy by improving the efficiency of simultaneous localisation and mapping through imitation learning. Lastly, we validate the advantages of behaviour trees' modularity in a real-world autonomous underwater vehicle's control system, and argue that this modularity contributes to efficiency, in terms of ease of use, and trust, in terms of facilitating human understanding.

Kritiska robotsystem är system vars funktion antingen är kritiska för slutförandet av en uppgift, eller kritiska på så sätt att ett misstag allvarligt kan skada människor eller miljö. Dessa kritiska aspekter fångas formellt av konvergens, i den meningen att systemets tillstånd går till en önskad region av tillståndsrummet, och säkerhet, i den meningen att systemets tillstånd undviker osäkra regioner i tillståndsrummet. Datadrivnakontrollpolicyer, hittade genom t.ex. imitationsinlärning eller förstärkningsinlärning, kan överträffa modellbaserade metoder för att effektivt uppnå konvergens och säkerhet; men de gör det ofta bara genom att öka möjligheterna för ett effektivt och säkert uppträdande, utan att ge några garantier, därför kan de vara svåra att lita på. Modellbaserade kontrollpolicyer, å andra sidan, är ofta väl lämpade för att möjliggöra formella garantier vad gäller konvergens och säkerhet, så de är ofta lättare att lita på. Huvudfrågan som ställs i denna avhandling är: hur kan vi kombinera datadrivna och modellbaserade styrpolicyer för att förbättra effektivitet samtidigt som vi formellt garanterar konvergens och säkerhet?

Vi besvarar denna fråga med Beteendeträd, ett ramverk för att representera hybridstyrsystem på ett modulärt sätt. Vi presenterar den första formella definitionen av beteendeträd som ett hybridsystem och presenterar villkoren under vilka exekveringen av ett beteendeträd som ett hybridkontrollsystem formellt kommer att garantera konvergens och säkerhet. Dessutom presenterar vi villkoren under vilka sådana formella garantier kan upprätthållas när man inkluderar overifierade datadrivna kontrollpolicyer, till exempel de som kommer från imitationsinlärning eller förstärkningsinlärning. Vi presenterar också ett tillvägagångssätt för att syntetisera sådana datadrivna kontrollpolicyer på ett sådant sätt att de stöttar konvergens och säkerhet genom att anpassa sig till oförutsedda händelser. Vid sidan av ovanstående utforskar vi också en viktig delfunktion inom robotautonomi genom att förbättra effektiviteten av samtidig lokalisering och kartläggning genom imitationsinlärning. Slutligen validerar vi fördelarna med behaviour trees modularitet i ett verkligt autonomt undervattensfordons kontrollsystem, och ser att denna modularitet bidrar till effektivitet, i termer av användarvänlighet och förtroende, när det gäller att underlätta mänsklig förståelse.

Gaussian processes (GPs) are becoming a standard tool to build terrain representations thanks to their capacity to model map uncertainty. This effectively yields a reliability measure of the areas of the map, which can be directly utilized by Bayes filtering algorithms in robot localization problems. A key factor is that this map uncertainty can incorporate the noise intrinsic to the terrain surveying process through the GPs ability to train on uncertain inputs (UIs). However, existing techniques to build GP maps with UIs in a tractable manner are restricted in the form and degree of the input noise. In this letter, we propose a flexible and efficient framework to build large-scale GP maps with UIs based on Stochastic Variational GPs and Monte Carlo sampling of the UIs distributions. We validate our mapping approach on a large bathymetric survey collected with an autonomous underwater vehicle (AUV) and analyze its performance against the use of deterministic inputs (DI). Finally, we show how using UI SVGP maps yields more accurate particle filter localization results than DI SVGP on a real AUV mission over an entirely predicted area.

Cyber-physical systems (CPSs) comprise a network of sensors and actuators that are integrated with a computing and communication core. Hydrobatic Autonomous Underwater Vehicles (AUVs) can be efficient and agile, offering new use cases in ocean production, environmental sensing and security. In this paper, a CPS concept for hydrobatic AUVs is validated in real-world field trials with the hydrobatic AUV SAM developed at the Swedish Maritime Robotics Center (SMaRC). We present system integration of hardware systems, software subsystems for mission planning using Neptus, mission execution using behavior trees, flight and trim control, navigation and dead reckoning. Together with the software systems, we show simulation environments in Simulink and Stonefish for virtual validation of the entire CPS. Extensive field validation of the different components of the CPS has been performed. Results of a field demonstration scenario involving the search and inspection of a submerged Mini Cooper using payload cameras on SAM in the Baltic Sea are presented. The full system including the mission planning interface, behavior tree, controllers, dead-reckoning and object detection algorithm is validated. The submerged target is successfully detected both in simulation and reality, and simulation tools show tight integration with target hardware.