Honda Research Institute: Aether
Sponsored by Honda Research Institute↗.
Facilitating innovative, real-world human-AI teaming research by using generative AI to streamline researcher tasks.
Responsibilities
I contributed to early-stage research and problem discovery. I also contributed heavily to ideation and wireframing of the final solution, focusing mainly on designing features to accelerate background research and building a component library and style guidelines. I also helped prepare final client-facing deliverables to be distributed through Honda Research Institute.
My Role
Product Design
UX Research
Duration
8 months
Team
3 Product Designers
1 UX Researcher
1 UX Engineer
Final Deliverables
High-Fidelity Prototype
Design System
Product Roadmap
Functional Website Prototype
Final Client Presentation
APPROACH
Work with what’s already known and be comfortable making assumptions when there is no clear direction forward.
There was no existing precedent for encouraging researchers to conduct more real-world studies. However, we realized we could still make big but reasonable assumptions while designing a system that both addresses current problems and tries to shape the future that the organization envisions.
BACKGROUND
Honda Research Institute is aiming to conduct mobility research in the real world.
As the organization shifts away from traditional vehicles and thinks more broadly about the future of mobility and transportation, it has realized it also needs to revisit how it currently performs research. Currently, most HRI research is done in lab or simulation environments, impeding the ability to account for spontaneous environmental variables.
PROBLEM
Researchers struggle to obtain study proposal approval because they are overwhelmed by the prerequisite tasks.
Before conducting early-stage research outside of the lab, researchers must undergo several preliminary activities, such as reading through prior literature and obtaining research approval, that are often time-consuming and tedious.
SOLUTION
Aether is an enterprise AI research assistant that streamlines researcher workflows so they can carry out activities faster.
AI-generated summaries enable effective analysis of relevant insights from past research.
Researchers can import papers to Aether, and then obtain comprehensive, organized summaries of the literature.
Annotating enables personalized organization of insights from research literature.
Memos allow researchers to make quick notes to themselves on the papers they’re reading.
Tags—either custom or AI-suggested—enable researchers to label excerpts with labels and retrieve them easily later.
AI-powered assistance accelerates drafting of review-ready study proposals.
Sometimes it’s hard coming up with tangible details that fit into the logistics. If researchers get stuck, they can draw ideas from AI recommendations to help them fill in the rest of the required research study information.
RESEARCHER PAIN POINTS
For researchers, getting started on a new project is the hardest part of the process.
Background research review and synthesis are tedious.
Researchers must go through A LOT (sometimes hundreds!) of previous papers to identify key findings to inform their own future studies.
There is little awareness about doing real-world research.
A general lack of awareness about how to carry out research in the real world compounded the first two pain points. Since most HRI researchers are accustomed to doing early-stage research in highly-controlled environments, they barely knew where they would begin if required to move their studies to real-life environments.
Combining details into a cohesive study plan is hard.
From methods to be used, to target participant demographics, everything must make sense as a whole for the study to be considered for approval.
IDEATION AND INITIAL CONCEPT
Envisioned a study planning guide that would “nudge” researchers into real-world studies, by providing assistance with the plan drafting.
We envisioned a smart research assistant that could reduce the time researchers spend trying to get their proposals approved. It would serve as a nudge to real-world research testing, to help researchers get started on a new proposal in an unfamiliar setting.
Web-like structure enables non-linear info input process for more flexibility.
A key insight we realized in the earlier stages of ideation was that planning a research study is a non-linear process. Instead of requiring users to fill out a certain module and then another, we made the template web-like so researchers could start and move in any direction.
Clicking into each module allows researchers to input details and get assistance from AI if needed.
After filling out modules, generate interactive study plan template to make more edits.
Researchers can input what they needed into different modules and receive AI-assisted suggestions or considerations as they drafted their study proposal.
For example, if the researcher plans to do the study in a location that requires special access, AI will point that out. The modules are for important details that need to be included in study planning, such as participant demographics, timeline, and research methods used.
TESTING FEEDBACK AND ITERATIONS
User testing revealed that researchers found the web layout a little perplexing; they said they would prefer a more straightforward visual organization. We changed the web structure to a simple top-to-bottom layout.
Changed the planning layout from a web to a vertical structure, as it was more intuitive.
Enabled plan draft updates in real-time; thought more broadly about using AI to “nudge” researchers.
In the first version, researchers could only see the draft research plan after completing the modules on the previous page. However, feedback indicated that it would be better if they could see how the plan changes with each module update, to reduce unnecessary back-and-forth navigation between pages.
Additionally, we began to think more broadly about how we could utilize AI to make the planning go by faster. Since most researchers don't know a lot about how to conduct low-fidelity testing in the real world, we decided to integrate generative AI so researchers could get suggestions to help them flesh out details for their plan.
Expanded the background research feature to address annotation and summarization needs during literature review.
In early iterations of Aether, we focused more on the planning portion than the background research one. However, later stakeholders mentioned that it would be helpful to have Aether also assist with organizing and annotating literature.
OUTCOMES
Improved HRI researcher productivity when performing crucial tasks.
Researchers found that they could complete tasks much faster while using Aether, and also liked that they could keep their papers organized in a centralized location.
“The ability to quickly annotate papers and extract key points is valuable.”
— Senior Research Scientist at HRI
Presented research findings, final product, and product roadmap to management.
We prepared an in-depth presentation to communicate to stakeholders and set the ball rolling on how they could realize their vision of becoming an AI and mobility research pioneer.
Additionally, we prepared a detailed client research report to be distributed throughout HRI.
REFLECTIONS
When dealing with a lot of ambiguity, keep asking “Why?”.
If I had a dollar every time we didn’t know what direction we were headed towards, I would’ve been able to pay off my tuition fees (maybe). There were so many times where it felt as if I was stumbling around blindfolded in a maze, but constantly asking “why” this project mattered and “why” it mattered that we try to facilitate real-world concept testing was crucial in navigating this future-facing project.
Don’t be afraid to make major changes! The design journey is messy and full of surprises.
We reframed our main problem A LOT of times throughout the project. It was scary at times to let go of our previous focus, but it was important that we trust ourselves and our insights to make new changes and pivot to new goals. This project truly was a rollercoaster and helped me accept that I don’t need to get things “right” the first time.
If we had more time, I would’ve liked to delve deeper into mitigating potential biases of our AI features.
We made sure to incorporate explainability into our Gen AI features, notably in AI recommendations made to researchers when they draft study plans. However, if time had permitted it would’ve been interesting to explore more proactive ways to make Aether more inclusive, such as including more diverse participant demographics.