Developing Virtual Assistants for the AEC-FM Industry

Our Motivation:

“Many BIM-based information systems have been developed for the AEC-FM industry. However, practitioners today still cannot realize the full benefits of rich, up-to-date information because of accessibility issues. This is because they must master not just the technical systems but also the digital information, language, and tools describing the systems.”

Research Contribution

Theoretically, this study will contribute a natural language (NL) -driven ontology model that supports an end-to-end semantic parser to translate complex NL utterances into system queries or functional calls.

Practically, it is expected to develop VA prototypes integrated with existing systems for use cases.

Problem

Practical Problem

The “three-hand problem” was formulated to illustrate how virtual assistants (VA) can help improve the efficiency of retrieving information about construction. 

Slow hand: Practitioners who don’t have the skills to use the systems need to wait for technical assistance to access information, resulting in decision latency. 

Difficult hand: It is difficult for practitioners to learn how to use new systems, because of lack of training resources, different backgrounds, and reluctance to change. 

Busy hand: Practitioners are often busy with their hands when performing tasks or wearing gloves. If they lack the needed information midway, it can result in reduction in productivity or poor quality of work.

Conceptual Problem

The AEC-FM industry lacked a NL-driven ontology that can map between NL query utterances and schemas in BIM-based systems.

Solution

To address the problem, the proposed VA can provide a natural language (NL) interface so that practitioners can directly “ask for” needed information from the systems with little learning.

NL is powerful in composing various filter conditions in queries and retaining contexts in the history of dialogues. As a result, users can spend minimal effort searching for the data.

Lastly, equipped with a speech recognition module, VA can make use of the voice to enable on-site and on-task information delivery in the hands-free condition.

Added Value For The Industry

By completion, this project will facilitate the development of VA for the AEC-FM industry, reducing the learning cost for systems and improving the efficiency of information retrieval as well as the user experience. In the long term, the NL interface will help bridge the gap between the white- and blue-collar workers in the AEC-FM industry by providing better access to technology.

Cooperation Partner

 Timeline

If you want to participate in the project please reach out to Junwen Zheng (junwenz@stanford.edu)