Danica Kragic

A central goal in sensory science is to establish quantitative mappings between physical stimuli and perceptual experience. Although such mappings are well defined in vision and audition, they remain elusive in olfaction, particularly for complex odor mixtures. Here, we show that perceptual distances between odor mixtures can be predicted with high fidelity and are unexpectedly well captured by a compact semantic space derived from single-molecule representations. In the Dialogue for Reverse Engineering Assessment and Methods Olfactory Mixtures Prediction Challenge, we assembled a unified dataset of odor-mixture pairs, benchmarked predictions on a hidden test set of 46 pairs, and integrated the top-performing models into a postchallenge ensemble. This model outperformed existing state-of-the-art approaches on the hidden test set, reducing RMSE by about 33% to 0.08 and increasing Pearson correlation by 53% to 0.57, and maintained strong performance on an independent validation set of 50 newly designed mixture pairs. An ensemble, retaining only olfactory semantic features for each model included, further improved predictions, raising the Pearson correlation by 7% to 0.61 on the test set and by 15% to 0.54 on the validation set. Given that semantic features were extracted from pure molecules, it suggests that mixture perception may not require fundamentally different representational principles from single-molecule olfaction. Together, these results establish a reproducible quantitative framework for olfactory mixture perception and advance efforts to measure, model, and engineer smell.

Transformer models pre-trained on sensory stimuli such as vision and audition have been shown to capture neural representations and to effectively predict brain responses to corresponding stimuli. Olfactory perception, however, remains a relatively underexplored sensory modality in this context. Understanding the alignment between chemical transformer models and brain activity during olfactory processing is particularly challenging due to the distributed nature of olfactory neural processing and the inherently subjective character of olfactory experience, which is strongly shaped by individual memory and prior exposure. In this work, we evaluate the representational alignment between pre-trained transformer models of chemical structure and functional magnetic resonance imaging recordings collected from human participants during olfactory perception. Using voxel-wise encoding models, we show that representations extracted from state-of-the-art molecular transformers exhibit strong alignment with olfactory brain responses, often exceeding handcrafted molecular fingerprints and physicochemical descriptors. We further analyze how this alignment varies across brain regions, model layers, and temporal response components, revealing distinct region-specific and temporal patterns. These findings represent a step toward understanding how that pre-trained chemical transformers capture meaningful structure relevant to human olfactory perception and provide insight into how molecular features relate to neural processing in the olfactory system.

1. Conversational agents (CAs) have become increasingly popular in applications such as survey studies (Kim et al., 2019; Ma et al., 2024; Xiao et al., 2020), education (Kumar et al., 2010; Winkler...

Foundation models offer a promising avenue to achieve transfer across robots, tasks, and environments.

Interactive artificial intelligence (AI) has rapidly emerged as a key field within both the human-computer interaction (HCI) and AI communities, driven by the growth of human-centered and responsible AI over the past few years. Unlike traditional AI approaches that prioritize algorithms, interactive AI emphasizes real-time manipulation, human-centered design, and transparency control in data and algorithms, with the overarching goal of enhancing human benefit. Simultaneously, AI-driven and embodied robotic systems are becoming increasingly integrated into daily life – homes, workplaces, public spaces, and healthcare – where human interaction is central to their success and acceptance. This makes it crucial to design robotic systems that are not only technically capable, but also understandable, trustworthy, and responsive to people’s needs, capabilities, and values. In light of this, we present the third edition of this InterAI workshop ¹ – Interactive AI for Human-Centered Robotics – to the CHI community, aiming to gain deeper insights of this emerging area through a half-day, in-person event. Our objective is to explore the current research landscape, identify challenges, and articulate future research directions for integrating interactive AI into human-centered robotic systems.

Humans naturally develop preferences for how manipulation tasks should be performed, which are often subtle, personal, and difficult to articulate. Although it is important for robots to account for these preferences to increase personalization and user satisfaction, they remain largely underexplored in robotic manipulation, particularly in the context of deformable objects like garments and fabrics. In this work, we study how to adapt pretrained visuomotor diffusion policies to reflect preferred behaviors using limited demonstrations. We introduce RKO, a novel preference-alignment method that combines the benefits of two recent frameworks: RPO and KTO. We evaluate RKO against common preference learning frameworks, including these two, as well as a baseline vanilla diffusion policy, on real-world cloth-folding tasks spanning multiple garments and preference settings. We show that preference-aligned policies (particularly RKO) achieve superior performance and sample efficiency compared to standard diffusion policy fine-tuning. These results highlight the importance and feasibility of structured preference learning for scaling personalized robot behavior in complex deformable object manipulation tasks.

Generalizing beyond the training domain in image-based behavior cloning remains challenging. Existing methods address individual axes of generalization, workspace shifts, viewpoint changes, and cross-embodiment transfer, yet they are typically developed in isolation and often rely on complex pipelines. We introduce PALM (Perception Alignment for Local Manipulation), which leverages the invariance of local action distributions between out-of-distribution (OOD) and demonstrated domains to address these OOD shifts concurrently, without additional input modalities, model changes, or data collection. PALM modularizes the manipulation policy into coarse global components and a local policy for fine-grained actions. We reduce the discrepancy between in-domain and OOD inputs at the local policy level by enforcing local visual focus and consistent proprioceptive representation, allowing the policy to retrieve invariant local actions under OOD conditions. Experiments show that PALM limits OOD performance drops to 8% in simulation and 24% in the real world, compared to 45% and 77% for baselines.