580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How do kinematic and multimodal cues from observed actions support human intention detection and social interaction?
This theme investigates how humans perceive intentions of others through observable physical cues such as movement kinematics, hand posture, eye gaze, and multimodal signals. Understanding these perceptual mechanisms is critical for modeling social cognition and improving human-robot interaction by enabling robots to infer or communicate intentions effectively.
Key finding: The study shows that reach-to-grasp kinematics are modulated by physical and social context, including object relations and eye-gaze communication, which serve as informative cues for detecting social intentions.... Read more
Key finding: Contrary to previous assumptions, adults were unable to reliably distinguish between grasp-to-drink and grasp-to-place intentions from naturalistic action kinematics embedded in richer visual scenes. Furthermore, motor... Read more
Key finding: This work presents a robust method for detecting users' intention to interact with a robot by fusing multimodal perceptual cues: head orientation, shoulder orientation, and vocal activity. Using RGB-D sensors and audio inputs... Read more
Key finding: The paper develops a task-level multimodal interaction model leveraging visual, auditory, tactile, and proxemic cues to establish and maintain human-robot engagement. By encoding both pragmatic and hedonic communication... Read more
Key finding: The study provides a computational framework under Active Inference explaining how neural circuits—particularly in the Posterior Parietal Cortex—compute flexible motor intentions to generate goal-directed actions in dynamic... Read more
2. What computational and data-driven models enable automated intention detection and recognition from behavioral or contextual data?
This theme focuses on algorithmic and machine learning approaches for recognizing user or agent intentions using data such as language, event logs, motion trajectories, or sensor streams. It addresses challenges in modeling latent intent states from observed behavior, leveraging probabilistic models, deep learning, and contextual knowledge to improve recognition accuracy, adaptability, and applicability in diverse domains including conversation systems, process mining, and assistive technologies.
Key finding: The authors derive an unsupervised computer vision algorithm that classifies agent behaviors as intentional or non-intentional solely based on 3D kinematics and fundamental physical principles like self-propelled and... Read more
Key finding: This paper proposes a context-aware Hidden Markov Model framework to mine user intentions from event log data, integrating contextual information that influences users' behavioral choices. The approach models intentions as... Read more
Key finding: The study introduces a novel approach for intent detection in natural language settings by transforming words into contextualized synset vectors using WordNet and deep bidirectional LSTM models. The method integrates semantic... Read more
Key finding: The paper presents a multi-agent system architecture to preprocess and structure unorganized activity logs for intention mining from large-scale event records, addressing data heterogeneity and noise challenges. By... Read more
Key finding: This review synthesizes state-of-the-art techniques for incremental intention recognition incorporating Bayesian Networks and logical reasoning to dynamically infer agents' goals based on observed actions and contextual... Read more
3. What theoretical distinctions and frameworks underpin the nature of intentions and their role in skilled action and communication?
This theme examines philosophical and cognitive science perspectives that clarify different types of intentions such as general versus practical, conditional versus unconditional intentions. It explores how intentions specify goals and contingency plans, interact with motor control architectures during skilled actions, and frame utterance interpretation and meaning. These conceptual insights provide foundational understanding informing computational modeling and empirical investigation of intentions.
Key finding: The paper distinguishes practical intentions—content-rich action-guiding control states—from general intentions, arguing that practical intentions specify means to achieve goals without directly controlling fine-grained... Read more
Key finding: The article contests the assumption that a speaker's actual communicative intention fully determines utterance meaning or that listeners aim to recover this intention. Instead, it argues that hearers focus on the speaker’s... Read more
Key finding: Through bibliometric analysis, this paper reveals that the concept of intention is rarely invoked explicitly in animal cognition studies and that human-centric frameworks are seldom adapted to non-human animals. Moreover, the... Read more