Agentic AI Research Team

Our team aims to develop AI technologies that integrate data-driven and knowledge-driven AI to understand the Semantics of human activities.
Human activities in daily living spaces are highly individualized and composed of various elements, including people, objects, and their interactions. Interpreting these highly individualized situations still depends heavily on the human who knows about the environment, tasks, and human activities. Providing this knowledge to an AI system to support daily life is still difficult.
For example, general-purpose robots or embodied-AI systems for home use require: 1) Knowledge of the physical characteristics of the individual living in the house and the indoor environment. 2) Knowledge of actions in performing the tasks (such as older adults support or young child care). 3) General knowledge of the human body, accident risk reports, and safety guidelines. 4) Service procedures. 5) Common knowledge of objects and events, etc.
The system must incorporate this multi-layered heterogeneous daily life knowledge into the inference and recognition process by linking it with information in different modalities, such as images and natural language cascading in multiple time and space occurrences.
To capture and integrate the multi-modality and heterogeneity nature of these data and knowledge, we conduct research on Knowledge Graphs, Machine Learning, Natural Language Processing, Embodied-AI, and Human-Agent-Interaction research.

DKIRT Youtube channel

■ DKIRT Youtube

Event Information

■ Daily Activity data generation in Cyberspace for Semantic AI technology and HRI simulation

We are hosting the Knowledge Graph Reasoning Challenge for Social Issues (KGRC4SI), where the task is to identify the risks in daily living activities to support the safety of older adults in their homes.The Challenge will provide two types of data sets generated by VirtualHome2KG, (1) the Video data recorded by a virtual space simulator reproducing daily life, and (2) Knowledge graphs representing human actions, object states, and relationships between scenes in the video

■ the event：https://challenge.knowledge-graph.jp/2022/ (In Japanese)
■ dataset: https://github.com/KnowledgeGraphJapan/KGRC-RDF/tree/kgrc4si(dataset labels are in English)

■ The 1st International Knowledge Graph Reasoning Challenge 2023

"In the challenge, the task is to estimate the culprits with a reasonable explanation using a dataset of knowledge graphs representing eight Sherlock Holmes mystery stories. "(cite: https://ikgrc.org/2023/)

■ the workshop: https://ikgrc.org/2023/
■ dataset：https://github.com/KnowledgeGraphJapan/KGRC-RDF/tree/ikgrc2023