Text evaluation has historically posed significant challenges, often demanding substantial labor and time cost. With the emergence of large language models (LLMs), researchers have explored LLMs' potential as alternatives for human evaluation. While these single-agent-based approaches show promise, experimental results suggest that further advancements are needed to bridge the gap between their current effectiveness and human-level evaluation quality. Recognizing that best practices of human evaluation processes often involve multiple human annotators collaborating in the evaluation, we resort to a multi-agent debate framework, moving beyond single-agent prompting strategies. The multi-agentbased approach enables a group of LLMs to synergize with an array of intelligent counterparts, harnessing their distinct capabilities and expertise to enhance efficiency and effectiveness in handling intricate tasks. In this paper, we construct a multi-agent referee team called ChatEval to autonomously discuss and evaluate the quality of generated responses from different models on open-ended questions and traditional natural language generation (NLG) tasks. We derive insights and lessons from practical scenarios where humans instigate group discussions for brainstorming and propose different communication strategies within ChatEval. Our experiments on two benchmark tasks illustrate that ChatEval delivers superior accuracy and correlation in alignment with human assessment. Furthermore, we find that the diverse role prompts (different personas) are essential in the multi-agent debate process; that is, utilizing the same role description in the prompt can lead to a degradation in performance. Our qualitative analysis also shows that ChatEval transcends mere textual scoring, offering a humanmimicking evaluation process for reliable assessments. Evaluating the quality of text generated by language models or written by humans has long been a challenging endeavor, consistently garnering substantial attention (Celikyilmaz et al., 2020). Traditional methodologies predominantly rely on human annotation of texts (Callison-Burch, 2009), an approach considered overly demanding in terms of time and cost.

Sequences of repeated gambles provide an experimental tool to characterize the risk preferences of humans or artificial decision-making agents. The difficulty of this inference depends on factors including the details of the gambles offered and the number of iterations of the game played. In this paper we explore in detail the practical challenges of inferring risk preferences from the observed choices of artificial agents who are presented with finite sequences of repeated gambles. We are motivated by the fact that the strategy to maximize long-run wealth for sequences of repeated additive gambles (where gains and losses are independent of current wealth) is different to the strategy for repeated multiplicative gambles (where gains and losses are proportional to current wealth.) Accurate measurement of risk preferences would be needed to tell whether an agent is employing the optimal strategy or not. To generalize the types of gambles our agents face we use the Yeo-Johnson transformation, a tool borrowed from feature engineering for time series analysis, to construct a family of gambles that interpolates smoothly between the additive and multiplicative cases. We then analyze the optimal strategy for this family, both analytically and numerically. We find that it becomes increasingly difficult to distinguish the risk preferences of agents as their wealth increases. This is because agents with different risk preferences eventually make the same decisions for sufficiently high wealth. We believe that these findings are informative for the effective design of experiments to measure risk preferences in humans.

The emergence of price comparison websites (PCWs) has presented insurers with unique challenges in formulating effective pricing strategies. Operating on PCWs requires insurers to strike a delicate balance between competitive premiums and profitability, amidst obstacles such as low historical conversion rates, limited visibility of competitors' actions, and a dynamic market environment. In addition to this, the capital intensive nature of the business means pricing below the risk levels of customers can result in solvency issues for the insurer. To address these challenges, this paper introduces reinforcement learning (RL) framework that learns the optimal pricing policy by integrating model-based and model-free methods. The model-based component is used to train agents in an offline setting, avoiding cold-start issues, while model-free algorithms are then employed in a contextual bandit (CB) manner to dynamically update the pricing policy to maximise the expected revenue. This facilitates quick adaptation to evolving market dynamics and enhances algorithm efficiency and decision interpretability. The paper also highlights the importance of evaluating pricing policies using an offline dataset in a consistent fashion and demonstrates the superiority of the proposed methodology over existing off-the-shelf RL/CB approaches. We validate our methodology using synthetic data, generated to reflect private commercially available data within real-world insurers, and compare against 6 other benchmark approaches. Our hybrid agent outperforms these benchmarks in terms of sample efficiency and cumulative reward with the exception of an agent that has access to perfect market information which would not be available in a real-world set-up.

Many models of learning in teams assume that team members can share solutions or learn concurrently. However, these assumptions break down in multidisciplinary teams where team members often complete distinct, interrelated pieces of larger tasks. Such contexts make it difficult for individuals to separate the performance effects of their own actions from the actions of interacting neighbors. In this work, we show that individuals can overcome this challenge by learning from network neighbors through mediating artifacts (like collective performance assessments). When neighbors' actions influence collective outcomes, teams with different networks perform relatively similarly to one another. However, varying a team's network can affect performance on tasks that weight individuals' contributions by network properties. Consequently, when individuals innovate (through ``exploring'' searches), dense networks hurt performance slightly by increasing uncertainty. In contrast, dense networks moderately help performance when individuals refine their work (through ``exploiting'' searches) by efficiently finding local optima. We also find that decentralization improves team performance across a battery of 34 tasks. Our results offer design principles for multidisciplinary teams within which other forms of learning prove more difficult.

This paper presents a new type of distributed dexterous manipulator: delta arrays. Our delta array setup consists of 64 linearly-actuated delta robots with 3D-printed compliant linkages. Through the design of the individual delta robots, the modular array structure, and distributed communication and control, we study a wide range of in-plane and out-of-plane manipulations, as well as prehensile manipulations among subsets of neighboring delta robots. We also demonstrate dexterous manipulation capabilities of the delta array using reinforcement learning while leveraging the compliance to not break the end-effectors. Our evaluations show that the resulting 192 DoF compliant robot is capable of performing various coordinated distributed manipulations of a variety of objects, including translation, alignment, prehensile squeezing, lifting, and grasping.

This paper presents an extension of the Mirror Descent method to overcome challenges in cooperative Multi-Agent Reinforcement Learning (MARL) settings, where agents have varying abilities and individual policies. The proposed Heterogeneous-Agent Mirror Descent Policy Optimization (HAMDPO) algorithm utilizes the multi-agent advantage decomposition lemma to enable efficient policy updates for each agent while ensuring overall performance improvements. By iteratively updating agent policies through an approximate solution of the trust-region problem, HAMDPO guarantees stability and improves performance. Moreover, the HAMDPO algorithm is capable of handling both continuous and discrete action spaces for heterogeneous agents in various MARL problems. We evaluate HAMDPO on Multi-Agent MuJoCo and StarCraftII tasks, demonstrating its superiority over state-of-the-art algorithms such as HATRPO and HAPPO. These results suggest that HAMDPO is a promising approach for solving cooperative MARL problems and could potentially be extended to address other challenging problems in the field of MARL.

The key innovation of our analytical method, CaRT, lies in establishing a new hierarchical, distributed architecture to guarantee the safety and robustness of a given learning-based motion planning policy. First, in a nominal setting, the analytical form of our CaRT safety filter formally ensures safe maneuvers of nonlinear multi-agent systems, optimally with minimal deviation from the learning-based policy. Second, in off-nominal settings, the analytical form of our CaRT robust filter optimally tracks the certified safe trajectory, generated by the previous layer in the hierarchy, the CaRT safety filter. We show using contraction theory that CaRT guarantees safety and the exponential boundedness of the trajectory tracking error, even under the presence of deterministic and stochastic disturbance. Also, the hierarchical nature of CaRT enables enhancing its robustness for safety just by its superior tracking to the certified safe trajectory, thereby making it suitable for off-nominal scenarios with large disturbances. This is a major distinction from conventional safety function-driven approaches, where the robustness originates from the stability of a safe set, which could pull the system over-conservatively to the interior of the safe set. Our log-barrier formulation in CaRT allows for its distributed implementation in multi-agent settings. We demonstrate the effectiveness of CaRT in several examples of nonlinear motion planning and control problems, including optimal, multi-spacecraft reconfiguration.

Mobile augmented reality (MAR) is widely acknowledged as one of the ubiquitous interfaces to the digital twin and Metaverse, demanding unparalleled levels of latency, computational power, and energy efficiency. The existing solutions for realizing MAR combine multiple technologies like edge, cloud computing, and fifth-generation (5G) networks. However, the inherent communication latency of visual data imposes apparent limitations on the quality of experience (QoE). To address the challenge, we propose an emergent semantic communication framework to learn the communication protocols in MAR. Specifically, we train two agents through a modified Lewis signaling game to emerge a discrete communication protocol spontaneously. Based on this protocol, two agents can communicate about the abstract idea of visual data through messages with extremely small data sizes in a noisy channel, which leads to message errors. To better simulate real-world scenarios, we incorporate channel uncertainty into our training process. Experiments have shown that the proposed scheme has better generalization on unseen objects than traditional object recognition used in MAR and can effectively enhance communication efficiency through the utilization of small-size messages.

Predicting pedestrians' trajectories is a crucial capability for autonomous vehicles' safe navigation, especially in spaces shared with pedestrians. Pedestrian motion in shared spaces is influenced by both the presence of vehicles and other pedestrians. Therefore, effectively modelling both pedestrian-pedestrian and pedestrian-vehicle interactions can increase the accuracy of the pedestrian trajectory prediction models. Despite the huge literature on ways to encode the effect of interacting agents on a pedestrian's predicted trajectory using deep-learning models, limited effort has been put into the effective selection of interacting agents. In the majority of cases, the interaction features used are mainly based on relative distances while paying less attention to the effect of the velocity and approaching direction in the interaction formulation. In this paper, we propose a heuristic-based process of selecting the interacting agents based on collision risk calculation. Focusing on interactions of potentially colliding agents with a target pedestrian, we propose the use of time-to-collision and the approach direction angle of two agents for encoding the interaction effect. This is done by introducing a novel polar collision grid map. Our results have shown predicted trajectories closer to the ground truth compared to existing methods (used as a baseline) on the HBS dataset.

Perspective-taking is the ability to perceive or understand a situation or concept from another individual's point of view, and is crucial in daily human interactions. Enabling robots to perform perspective-taking remains an unsolved problem; existing approaches that use deterministic or handcrafted methods are unable to accurately account for uncertainty in partially-observable settings. This work proposes to address this limitation via a deep world model that enables a robot to perform both perception and conceptual perspective taking, i.e., the robot is able to infer what a human sees and believes. The key innovation is a decomposed multi-modal latent state space model able to generate and augment fictitious observations/emissions. Optimizing the ELBO that arises from this probabilistic graphical model enables the learning of uncertainty in latent space, which facilitates uncertainty estimation from high-dimensional observations. We tasked our model to predict human observations and beliefs on three partially-observable HRI tasks. Experiments show that our method significantly outperforms existing baselines and is able to infer visual observations available to other agent and their internal beliefs.