Embedded Files
Why do robots that work in controlled settings fail in the real world?
Why do robots that work in controlled settings fail in the real world?
Robots have proven themselves in factories and warehouses, but adoption becomes harder in open-ended environments: construction sites, city streets, hospitals, and homes. These settings are messy, high-risk, and full of human judgment. Our question was simple: when robots are technically capable, why do they still struggle to scale?
What I did
What I did
As part of a grant-funded, multi-year research program with Northeastern University’s Center for Experiential Robotics, we conducted 30+ semi-structured interviews with robotics founders, technical leads, investors, industry partners, and government experts. I led the construction adoption stream, combining interviews, field engagement, and comparative analysis of robotaxis. Using PitchBook and public company data, I built and categorized a landscape of 300+ construction robotics companies, then translated that analysis into a structured opportunity-mapping workshop where industry participants identified existing solutions and unmet automation needs across the construction lifecycle.
Field research: Site visit at Reframe Systems, an automated modular construction startup.
Industry engagement: Suffolk Technologies’ BOOST Demo Day in Boston.
Facilitated discovery: Opportunity-mapping workshop with researchers and industry participants.
What we found
What we found
The barrier is rarely the robot. Success depends on the setting: fragmented workflows, real-time judgment calls, safety expectations, liability, and trust can defeat technically capable machines.
Resistance is often perceptual, not only factual. Adoption depends on whether users, buyers, and regulators believe the system can be trusted in edge cases, not just whether the technology works in demos.
Narrow use cases travel better. Broad automation stories often fail because they ask organizations to change too much at once. Successful robotics adoption starts with a specific task, a clear buyer, and a low-resistance entry point. (E.g., Katerra raised over $2 billion to automate construction end-to-end, and went bankrupt. Rugged Robotics automated one task, layout marking, and won.)
Edge-case learning becomes the moat. Companies that accumulate evidence from rare, messy situations build a compounding advantage that competitors cannot easily shortcut.
To put these findings in context, we mapped robotics applications across two conditions that shape the adoption environment: the potential for human harm and the frequency of human encounters.
A contextual map of robotics adoption environments.
So what
So what
We distilled these findings into a two-stage playbook for launching and scaling robots in messy environments: first land on a narrow, low-resistance use case; then scale by making the surrounding system more robot-ready through stakeholder education, workflow redesign, and accumulated domain evidence.
To activate the research beyond the paper, I designed and moderated a 100-attendee Robotics x Construction Demo Day with 50+ industry participants, including robotics startups, enterprise buyers, investors, and researchers. The event used demos and facilitated discussion to test how buyers evaluate new robotics solutions, surface adoption barriers, and generate follow-up pilot and vendor-evaluation conversations.
From evidence to strategy
From evidence to strategy
The resulting playbook translates field evidence into practical guidance for where to start, how to earn trust, and how to scale through stakeholder learning and domain data.
A two-stage playbook for scaling robots in messy environments.
With Fernando F. Suarez and Nathan Rietzler, Northeastern University. Working paper, 2026.
Page updated
Google Sites
Report abuse