Sortition, the random selection of political representatives, is increasingly being used around the world to choose participants of deliberative processes like Citizens' Assemblies. Motivated by sortition's practical importance, there has been a recent flurry of research on sortition algorithms, whose task it is to select a panel from among a pool of volunteers. This panel must satisfy quotas enforcing representation of key population subgroups. Past work has contributed an algorithmic approach for fulfilling this task while ensuring that volunteers' chances of selection are maximally equal, as measured by any convex equality objective. The question, then, is: which equality objective is the right one? Past work has mainly studied the objectives Minimax and Leximin, which respectively minimize the maximum and maximize the minimum chance of selection given to any volunteer. Recent work showed that both of these objectives have key weaknesses: Minimax is highly robust to manipulation but is arbitrarily unfair; oppositely, Leximin is highly fair but arbitrarily manipulable. In light of this gap, we propose a new equality objective, Goldilocks, that aims to achieve these ideals simultaneously by ensuring that no volunteer receives too little or too much chance of selection. We theoretically bound the extent to which Goldilocks achieves these ideals, finding that in an important sense, Goldilocks recovers among the best available solutions in a given instance. We then extend our bounds to the case where the output of Goldilocks is transformed to achieve a third goal, Transparency. Our empirical analysis of Goldilocks in real data is even more promising: we find that this objective achieves nearly instance-optimal minimum and maximum selection probabilities simultaneously in most real instances -- an outcome not even guaranteed to be possible for any algorithm.

Large Language Models (LLMs) have shown exceptional results on current benchmarks when working individually. The advancement in their capabilities, along with a reduction in parameter size and inference times, has facilitated the use of these models as agents, enabling interactions among multiple models to execute complex tasks. Such collaborations offer several advantages, including the use of specialized models (e.g. coding), improved confidence through multiple computations, and enhanced divergent thinking, leading to more diverse outputs. Thus, the collaborative use of language models is expected to grow significantly in the coming years. In this work, we evaluate the behavior of a network of models collaborating through debate under the influence of an adversary. We introduce pertinent metrics to assess the adversary's effectiveness, focusing on system accuracy and model agreement. Our findings highlight the importance of a model's persuasive ability in influencing others. Additionally, we explore inference-time methods to generate more compelling arguments and evaluate the potential of prompt-based mitigation as a defensive strategy.

Cooperative Multi-Agent Reinforcement Learning (CMARL) strategies are well known to be vulnerable to adversarial perturbations. Previous works on adversarial attacks have primarily focused on white-box attacks that directly perturb the states or actions of victim agents, often in scenarios with a limited number of attacks. However, gaining complete access to victim agents in real-world environments is exceedingly difficult. To create more realistic adversarial attacks, we introduce a novel method that involves injecting traitor agents into the CMARL system. We model this problem as a Traitor Markov Decision Process (TMDP), where traitors cannot directly attack the victim agents but can influence their formation or positioning through collisions. In TMDP, traitors are trained using the same MARL algorithm as the victim agents, with their reward function set as the negative of the victim agents' reward. Despite this, the training efficiency for traitors remains low because it is challenging for them to directly associate their actions with the victim agents' rewards. To address this issue, we propose the Curiosity-Driven Adversarial Attack (CuDA2) framework. CuDA2 enhances the efficiency and aggressiveness of attacks on the specified victim agents' policies while maintaining the optimal policy invariance of the traitors. Specifically, we employ a pre-trained Random Network Distillation (RND) module, where the extra reward generated by the RND module encourages traitors to explore states unencountered by the victim agents. Extensive experiments on various scenarios from SMAC demonstrate that our CuDA2 framework offers comparable or superior adversarial attack capabilities compared to other baselines.

In order to ensure autonomous vehicles are safe for on-road deployment, simulation-based testing has become an integral complement to on-road testing. The rise in simulation testing and validation reflects a growing need to verify that AV behavior is consistent with desired outcomes even in edge case scenarios $-$ which may seldom or never appear in on-road testing data. This raises a critical question: to what extent are AV failures in simulation consistent with data collected from real-world testing? As a result of the gap between simulated and real sensor data (sim-to-real gap), failures in simulation can either be spurious (simulation- or simulator-specific issues) or relevant (safety-critical AV system issues). One possible method for validating if simulated time series failures are consistent with real world time series sensor data could involve retrieving instances of the failure scenario from a real-world time series dataset, in order to understand AV performance in these scenarios. Adopting this strategy, we propose a formal definition of what constitutes a match between a real-world labeled time series data item and a simulated scenario written from a fragment of the Scenic probabilistic programming language for simulation generation. With this definition of a match, we develop a querying algorithm that identifies the subset of a labeled time series dataset matching a given scenario. To allow this approach to be used to verify the safety of other cyber-physical systems (CPS), we present a definition and algorithm for matching scalable beyond the autonomous vehicles domain. Experiments demonstrate the precision and scalability of the algorithm for a set of challenging and uncommon time series scenarios identified from the nuScenes autonomous driving dataset. We include a full system implementation of the querying algorithm freely available for use across a wide range of CPS.

Electromechanical systems manage physical processes through a network of inter-connected components. Today, programming the interactions required for coordinating these components is largely a manual process. This process is time-consuming and requires manual adaptation when system features change. To overcome this issue, we use autonomous software agents that process semantic descriptions of the system to determine coordination requirements and constraints; on this basis, they then interact with one another to control the system in a decentralized and coordinated manner.Our core insight is that coordination requirements between individual components are, ultimately, largely due to underlying physical interdependencies between the components, which can be (and, in many cases, already are) semantically modeled in automation projects. Agents then use hypermedia to discover, at run time, the plans and protocols required for enacting the coordination. A key novelty of our approach is the use of hypermedia-driven interaction: it reduces coupling in the system and enables its run-time adaptation as features change.

Battery-less embedded devices powered by energy harvesting are increasingly being used in wireless sensing applications. However, their limited and often uncertain energy availability challenges designing application programs. To examine if BDI-based agent programming can address this challenge, we used it for a real-life application involving an environmental sensor that works on energy harvested from ambient light. This yielded the first ever implementation of a BDI agent on a low-power battery-less and energy-harvesting embedded system. Furthermore, it uncovered conceptual integration challenges between embedded systems and BDI-based agent programming that, if overcome, will simplify the deployment of more autonomous systems on low-power devices with non-deterministic energy availability. Specifically, we (1) mapped essential device states to default \textit{internal} beliefs, (2) recognized and addressed the need for beliefs in general to be \textit{short-} or \textit{long-term}, and (3) propose dynamic annotation of intentions with their run-time energy impact. We show that incorporating these extensions not only simplified the programming but also improved code readability and understanding of its behavior.

We introduce the Overcooked Generalisation Challenge (OGC) - the first benchmark to study agents' zero-shot cooperation abilities when faced with novel partners and levels in the Overcooked-AI environment. This perspective starkly contrasts a large body of previous work that has trained and evaluated cooperating agents only on the same level, failing to capture generalisation abilities required for real-world human-AI cooperation. Our challenge interfaces with state-of-the-art dual curriculum design (DCD) methods to generate auto-curricula for training general agents in Overcooked. It is the first cooperative multi-agent environment specially designed for DCD methods and, consequently, the first benchmarked with state-of-the-art methods. It is fully GPU-accelerated, built on the DCD benchmark suite minimax, and freely available under an open-source license: https://git.hcics.simtech.uni-stuttgart.de/public-projects/OGC. We show that current DCD algorithms struggle to produce useful policies in this novel challenge, even if combined with recent network architectures that were designed for scalability and generalisability. The OGC pushes the boundaries of real-world human-AI cooperation by enabling the research community to study the impact of generalisation on cooperating agents.

We propose a control synthesis framework for a heterogeneous multi-robot system to satisfy collaborative tasks, where actions may take varying duration of time to complete. We encode tasks using the discrete logic LTL^\psi, which uses the concept of bindings to interleave robot actions and express information about relationship between specific task requirements and robot assignments. We present a synthesis approach to automatically generate a teaming assignment and corresponding discrete behavior that is correct-by-construction for continuous execution, while also implementing synchronization policies to ensure collaborative portions of the task are satisfied. We demonstrate our approach on a physical multi-robot system.

Active Voltage Control (AVC) on the Power Distribution Networks (PDNs) aims to stabilize the voltage levels to ensure efficient and reliable operation of power systems. With the increasing integration of distributed energy resources, recent efforts have explored employing multi-agent reinforcement learning (MARL) techniques to realize effective AVC. Existing methods mainly focus on the acquisition of short-term AVC strategies, i.e., only learning AVC within the short-term training trajectories of a singular diurnal cycle. However, due to the dynamic nature of load demands and renewable energy, the operation states of real-world PDNs may exhibit significant distribution shifts across varying timescales (e.g., daily and seasonal changes). This can render those short-term strategies suboptimal or even obsolete when performing continuous AVC over extended periods. In this paper, we propose a novel temporal prototype-aware learning method, abbreviated as TPA, to learn time-adaptive AVC under short-term training trajectories. At the heart of TPA are two complementary components, namely multi-scale dynamic encoder and temporal prototype-aware policy, that can be readily incorporated into various MARL methods. The former component integrates a stacked transformer network to learn underlying temporal dependencies at different timescales of the PDNs, while the latter implements a learnable prototype matching mechanism to construct a dedicated AVC policy that can dynamically adapt to the evolving operation states. Experimental results on the AVC benchmark with different PDN sizes demonstrate that the proposed TPA surpasses the state-of-the-art counterparts not only in terms of control performance but also by offering model transferability. Our code is available at https://github.com/Canyizl/TPA-for-AVC.

Learning generalist embodied agents, able to solve multitudes of tasks in different domains is a long-standing problem. Reinforcement learning (RL) is hard to scale up as it requires a complex reward design for each task. In contrast, language can specify tasks in a more natural way. Current foundation vision-language models (VLMs) generally require fine-tuning or other adaptations to be functional, due to the significant domain gap. However, the lack of multimodal data in such domains represents an obstacle toward developing foundation models for embodied applications. In this work, we overcome these problems by presenting multimodal foundation world models, able to connect and align the representation of foundation VLMs with the latent space of generative world models for RL, without any language annotations. The resulting agent learning framework, GenRL, allows one to specify tasks through vision and/or language prompts, ground them in the embodied domain's dynamics, and learns the corresponding behaviors in imagination. As assessed through large-scale multi-task benchmarking, GenRL exhibits strong multi-task generalization performance in several locomotion and manipulation domains. Furthermore, by introducing a data-free RL strategy, it lays the groundwork for foundation model-based RL for generalist embodied agents.