We construct a robust and diverse benchmark test set to comprehensively evaluate language models in the context of natural language interactions with non-player characters (NPCs) for role-playing games. Such models need to be able to stay in character, stay consistent with the game world, and answer appropriately when confronted with real-world events and entities which the character should not be aware of (e.g., cars or Taylor Swift for a medieval setting). In addition, such characters need to be safe, avoiding harmful or toxic content. Results indicate that our novel prompting strategy led to improved performance on this new benchmark, for example showing a 6 https://github.com/WabbaMan/npc-bench .

We investigate the use of generative AI to create complex speech-enabled conversational characters for immersive and safe interactive narratives in video games. To explore this, we develop a murder-mystery game in Minecraft, featuring large language model (LLM)-driven non-player characters (NPCs) that support open-ended, voice-based interaction. By combining GPT-4o with ASR and TTS modules, players can converse with characters, and attempt to solve a mystery through natural dialogue. Prompt engineering techniques are applied to enhance character consistency, safety, and narrative depth. Evaluations show improved performance in terms of character correctness, dialogue coherence, and safety – for example improving safety and character consistency significantly (p = 0.00195). The project materials are available at [github] . Demo video available at: [video] .

We investigate whether automatic speech recognition (ASR) errors in user input can lead to unsafe outputs from Embodied AI (EAI) models. We find that ASR errors can lead to harmful instructions being accepted and executed by EAI models, thereby reducing safety. We simulate ASR errors and combine them with existing safety benchmarks (SafeAgentBench and POEX) to evaluate how different errors affect embodied AI safety. We find that some of them preserve semantic structure but increase harmful ambiguity, while others weaken the model refusal behaviour and allow unsafe plans to be generated and executed. We show that in some cases automatic correction of ASR errors can reduce the risk, but this is not always effective. Overall, we show that ASR errors lead to significant safety risks for embodied AI.

The integration of Large Language Models (LLMs) into multiagent systems has opened new possibilities for collaborative reasoning and cooperation with AI agents. This paper explores different prompting methods and evaluates their effectiveness in enhancing agent collaborative behaviour and decision-making. We enhance CoELA, a framework designed for building Collaborative Embodied Agents that leverage LLMs for multi-agent communication, reasoning, and task coordination in shared virtual spaces. Through systematic experimentation, we examine different LLMs and prompt engineering strategies to identify optimised combinations that maximise collaboration performance. Furthermore, we extend our research by integrating speech capabilities, enabling seamless collaborative voice-based interactions. Our findings highlight the effectiveness of prompt optimisation in enhancing collaborative agent performance; for example, our best combination improved the efficiency of the system running with Gemma3 by 22 Alice and Bob – Voice Chat GUI Video demonstration . Github link to codebase .

Managing conversational interactions with groups of people is still an open challenge in human-robot interaction, requiring a multi-modal combination of sensory inputs/outputs and dialogue systems. In this paper, we present the development of an integrated multi-modal system connecting a Large Language Model (LLM) with a social robot's perception and action modules for managing situated multi-party interactions. We describe and discuss the exploratory results of a system-wide performance evaluation via a within-subjects user study in which 27 unique pairs of participants interacted with a social robot under two conditions: a multi-party capable system and a baseline system with only single-party capabilities. Participants interacted with the two systems in a combination of task-based and open-ended scenarios, for a total of 108 interactions with each of the two systems. Our evaluation demonstrated a slight preference for the Multi-Party system and a more balanced interaction overall, and highlights potentials and open challenges in the integration of LLMs capabilities into robotic conversational systems.

We examine three evaluation paradigms: standard benchmarks (e.g., MMLU and BBH), interactive games (e.g., Signalling Games or Taboo), and cognitive tests (e.g., for working memory or theory of mind). First, we investigate which of the former two-benchmarks or games-is most effective at discriminating LLMs of varying quality. Then, inspired by human cognitive assessments, we compile a suite of targeted tests that measure cognitive abilities deemed essential for effective language use, and we investigate their correlation with model performance in benchmarks and games. Our analyses reveal that interactive games are superior to standard benchmarks in discriminating models. Causal and logical reasoning correlate with both static and interactive tests, while differences emerge regarding core executive functions and social/emotional skills, which correlate more with games. We advocate for the development of new interactive benchmarks and targeted cognitive tasks inspired by assessing human abilities but designed specifically for LLMs.

Despite the many recent achievements in developing and deploying social robotics, there are still many underexplored environments and applications for which systematic evaluation of such systems by end-users is necessary. While several robotic platforms have been used in gerontological healthcare, the question of whether or not a social interactive robot with multi-modal conversational capabilities will be useful and accepted in real-life facilities is yet to be answered. This paper is an attempt to partially answer this question, via two waves of experiments with patients and companions in a day-care gerontological facility in Paris with a full-sized humanoid robot endowed with social and conversational interaction capabilities. The software architecture, developed during the H2020 SPRING project, together with the experimental protocol, allowed us to evaluate the acceptability (AES) and usability (SUS) with more than 60 end-users. Overall, the users are receptive to this technology, especially when the robot perception and action skills are robust to environmental clutter and flexible to handle a plethora of different interactions.

Machine learning has exhibited substantial success in the field of natural language processing (NLP). For example, large language models have empirically proven to be capable of producing text of high complexity and cohesion. However, at the same time, they are prone to inaccuracies and hallucinations. As these systems are increasingly integrated into real-world applications, ensuring their safety and reliability becomes a primary concern. There are safety critical contexts where such models must be robust to variability or attack and give guarantees over their output. Computer vision had pioneered the use of formal verification of neural networks for such scenarios and developed common verification standards and pipelines, leveraging precise formal reasoning about geometric properties of data manifolds. In contrast, NLP verification methods have only recently appeared in the literature. While presenting sophisticated algorithms in their own right, these papers have not yet crystallised into a common methodology. They are often light on the pragmatical issues of NLP verification, and the area remains fragmented. In this paper, we attempt to distil and evaluate general components of an NLP verification pipeline that emerges from the progress in the field to date. Our contributions are twofold. First, we propose a general methodology to analyse the effect of the embedding gap – a problem that refers to the discrepancy between verification of geometric subspaces, and the semantic meaning of sentences which the geometric subspaces are supposed to represent. We propose a number of practical NLP methods that can help to quantify the effects of the embedding gap. Second, we give a general method for training and verification of neural networks that leverages a more precise geometric estimation of semantic similarity of sentences in the embedding space and helps to overcome the effects of the embedding gap in practice.

Interaction between learner and feedback-giver has come into focus recently for post-training of Large Language Models (LLMs), through the use of reward models that judge the appropriateness of a model's response. In this paper, we investigate whether Dialogue Games – goal-directed and rule-governed activities driven predominantly by verbal actions – can also serve as a source of feedback signals for learning. We introduce Playpen, an environment for off- and online learning through Dialogue Game self-play, and investigate a representative set of post-training methods: supervised fine-tuning; direct alignment (DPO); and reinforcement learning with GRPO. We experiment with post-training a small LLM (Llama-3.1-8B-Instruct), evaluating performance on unseen instances of training games as well as unseen games, and on standard benchmarks. We find that imitation learning through SFT improves performance on unseen instances, but negatively impacts other skills, while interactive learning with GRPO shows balanced improvements without loss of skills. We release the framework and the baseline training setups to foster research in the promising new direction of learning in (synthetic) interaction.

Interaction between learner and feedback-giver has come into focus recently for post-training of Large Language Models (LLMs), through the use of reward models that judge the appropriateness of a model’s response. In this paper, we investigate whether Dialogue Games—goal-directed and rule-governed activities driven predominantly by verbal actions—can also serve as a source of feedback signals for learning.We introduce Playpen, an environment for off- and online learning through Dialogue Game self-play, and investigate a representative set of post-training methods: supervised fine-tuning; direct alignment (DPO); and reinforcement learning with Group Relative Policy Optimization (GRPO). We experiment with post-training a small LLM (Llama-3.1-8B-Instruct), evaluating performance on unseen instances of training games as well as unseen games, and on standard benchmarks. We find that imitation learning through SFT improves performance on unseen instances, but negatively impacts other skills, while interactive learning with GRPO shows balanced improvements without loss of skills. We release the framework and the baseline training setups to foster research in this promising new direction of “learning in (synthetic) interaction”.

This study explores replacing Transformers in Visual Language Models (VLMs) with Mamba, a recent structured state space model (SSM) that demonstrates promising performance in sequence modeling. We test models up to 3B parameters under controlled conditions, showing that Mamba-based VLMs outperforms Transformers-based VLMs in captioning, question answering, and reading comprehension. However, we find that Transformers achieve greater performance in visual grounding and the performance gap widens with scale. We explore two hypotheses to explain this phenomenon: 1) the effect of task-agnostic visual encoding on the updates of the hidden states, and 2) the difficulty in performing visual grounding from the perspective of in-context multimodal retrieval. Our results indicate that a task-aware encoding yields minimal performance gains on grounding, however, Transformers significantly outperform Mamba at in-context multimodal retrieval. Overall, Mamba shows promising performance on tasks where the correct output relies on a summary of the image but struggles when retrieval of explicit information from the context is required.

An effective method for combining frozen large language models (LLM) and visual encoders involves a resampler module that creates a 'visual prompt' which is provided to the LLM, along with the textual prompt. While this approach has enabled impressive performance across many coarse-grained tasks like image captioning and visual question answering, (Alayrac et al., 2022; Dai et al., 2023), more fine-grained tasks that require spatial understanding have not been thoroughly examined. In this paper, we use diagnostic classifiers to measure the extent to which the visual prompt produced by the resampler encodes spatial information. Our results show that this information is largely absent from the resampler output when kept frozen during training of the classifiers. However, when the resampler and classifier are trained jointly, we observe a significant performance boost. This shows that the compression achieved by the resamplers can in principle encode the requisite spatial information, but that more object-aware objectives are needed at the pretraining stage to facilitate this capability(1).

We have deployed an LLM-based spoken dialogue system in a real hospital. The ARI social robot embodies our system, which patients and their companions can have multi-party conversations with together. In order to enable this multi-party ability, multimodality is critical. Our system, therefore, receives speech and video as input, and generates both speech and gestures (arm, head, and eye movements). In this paper, we describe our complex setting and the architecture of our dialogue system. Each component is detailed, and a video of the full system is available with the appropriate components highlighted in real-time. Our system decides when it should take its turn, generates human-like clarification requests when the patient pauses mid-utterance, answers in-domain questions (grounding to the in-prompt knowledge), and responds appropriately to out-of-domain requests (like generating jokes or quizzes). This latter feature is particularly remarkable as real patients often utter unexpected sentences that could not be handled previously.

In recent years, several machine learning models have been proposed. They are trained with a language modelling objective on large-scale text-only data. With such pretraining, they can achieve impressive results on many Natural Language Understanding and Generation tasks. However, many facets of meaning cannot be learned by “listening to the radio” only. In the literature, many Vision+Language (V+L) tasks have been defined with the aim of creating models that can ground symbols in the visual modality. In this work, we provide a systematic literature review of several tasks and models proposed in the V+L field. We rely on Wittgenstein’s idea of ‘language games’ to categorise such tasks into 3 different families: 1) discriminative games, 2) generative games, and 3) interactive games. Our analysis of the literature provides evidence that future work should be focusing on interactive games where communication in Natural Language is important to resolve ambiguities about object referents and action plans and that physical embodiment is essential to understand the semantics of situations and events. Overall, these represent key requirements for developing grounded meanings in neural models.

The Human - Large Language Model Interaction workshop aims at bringing together researchers and industry experts across all disciplines in the fields of Human-Robot Interaction and Artificial Intelligence for an interactive and interdisciplinary discussion around the enormous opportunities and challenges that emerge from integrating Large Language Models in the interactive, conversational and reasoning abilities of robots. We will discuss the impact that these models will increasingly have on the way that humans and robots interact by focusing on the limitations, challenges, risks and opportunities that such models represent and how can they best be addressed by the HRI community. The workshop aims to provide an important venue to encourage debate around issues concerning the deployment of Large Language Models empowered solutions for human-robot interaction. This will be an opportunity to share and discuss ideas, worries, strategies, insights and findings around the application of Large Language Models in interaction scenarios. We hope these discussions will help to inform and guide the community towards the design, development and implementation of safe, ethical, and responsible 'new breed' of social robots.

The Generalized Referring Expression Comprehension (GREC) task extends classic REC by generating image bounding boxes for objects referred to in natural language expressions, which may indicate zero, one, or multiple targets. This generalization enhances the practicality of REC models for diverse real-world applications. However, the presence of varying numbers of targets in samples makes GREC a more complex task, both in terms of training supervision and final prediction selection strategy. Addressing these challenges, we introduce RECANTFormer, a one-stage method for GREC that combines a decoder-free (encoder-only) transformer architecture with DETR-like Hungarian matching. Our approach consistently outperforms baselines by significant margins in three GREC datasets.

While research in multi-party spoken conversation with intelligent embodied agents has made significant progress in sub-tasks like speaker identification and non-verbal cues, there’s a gap in fully autonomous applications users can directly interact with. This lack translates to the absence of a standard methodology for evaluating multi-party conversational speech agents that considers both task-based system performance and user experience. Our research has addressed the former by developing a multi-modal robot receptionist for a hospital waiting room whose multi-party conversational ability, nonverbal behaviour, and dialogue management is implemented using Large Language Models (LLM). In this paper, we go on to address the issue of evaluation, describing an experimental methodology and design of task-based user experiments that captures both objective measures of multi-party dialogue performance (such as accurate tracking of user goals) and the users’ subjective experience of multi-party embodied conversations. This paper therefore presents a holistic methodology for the future evaluation of multi-party spoken conversational agents.

AI personal assistants deployed via robots or wearables require embodied understanding to collaborate with humans effectively. However, current Vision-Language Models (VLMs) primarily focus on third-person view videos, neglecting the richness of egocentric perceptual experience. To address this gap, we propose three key contributions. First, we introduce the Egocentric Video Understanding Dataset (EVUD) for training VLMs on video captioning and question answering tasks specific to egocentric videos. Second, we present , a 7B parameter VLM trained using parameter-efficient methods on EVUD. Finally, we evaluate ‘s capabilities on OpenEQA, a challenging benchmark for embodied video question answering. Our model achieves state-of-the-art performance, outperforming open-source models including strong Socratic models using GPT-4 as a planner by 3.6%.Additionally, we outperform Claude 3 and Gemini Pro Vision 1.0 and showcase competitive results compared to Gemini Pro 1.5 and GPT-4V, even surpassing the latter in spatial reasoning. This research paves the way for building efficient VLMs that can be deployed in robots or wearables, leveraging embodied video understanding to collaborate seamlessly with humans in everyday tasks, contributing to the advancement of next-generation Embodied AI.