If you know my views on autonomous systems in construction, you know I'm not betting on full autonomy winning the near term. I'm a proponent of teleoperation — human-operated systems deployed with remote control capabilities. That position comes from years of thinking about how complex machinery actually gets adopted across industries like construction and manufacturing, and it's grounded in three realities that don't get discussed enough: insurance, liability, and regulatory accountability.
In a fully autonomous system, when something goes wrong, liability fragments across the vehicle manufacturer, the software developer, the sensor supplier, and potentially the site owner — a legal web that no insurance framework has cleanly resolved. The autonomous excavator liability insurance market only reached $1.18 billion in 2024 — a fraction of the overall equipment market — reflecting how early and unsettled the risk frameworks still are. Teleoperation sidesteps this cleanly: there is a human in the loop, there is a clear accountable operator, and existing liability structures apply. That's not a limitation — it's a feature. It's why teleoperated systems are being deployed at scale right now while fully autonomous ones are still in pilots.
This article is about what comes next for teleoperation. Not the machine side — that problem is largely solved. The missing layer is the operator.
We've spent years solving the machine side of remote operation
Latency is down. Bandwidth is up. Camera angles are better. Haptic feedback is improving. Teleoperated excavators, cranes, and dozers are real, robust, and deployed at scale — from mining sites in Australia to hazmat zones in Europe. The equipment is no longer the bottleneck. The operator is.
Construction equipment accidents frequently result from prolonged operator inattention, leading to serious injuries and fatalities on site. Research shows that fatigue alone explains up to 37% of the variance in a worker's ability to identify hazards. And yet, every teleoperation system on the market today treats the human in the loop as a black box.
We monitor the machine constantly. Fuel levels. Engine temperature. Hydraulic pressure. Cycle times. AI-enabled predictive maintenance has already reduced unplanned equipment downtime by 25% for major rental fleets. The telematics market for construction equipment is projected to grow from $2.8 billion in 2025 to $9 billion by 2035. (Source: Global Market Insights, 2026)
But the operator? We give them a shift schedule and a coffee.
For decades, EEG data was treated as a clinical tool — useful in hospital settings, impractical everywhere else. Signals were noisy, highly variable across individuals, and required extensive per-person calibration before they could be meaningfully interpreted. That constraint is now breaking down fast.
The most advanced EEG foundation models are now trained on over 8 million segments, achieving robust brain signal decoding across tasks and subjects — the ST-EEGFormer model won first place at the NeurIPS 2025 EEG Challenge and underpins an ICLR 2026 benchmark study. Models like the Large Cognition Model use large-scale self-supervised learning to capture universal EEG representations, enabling efficient fine-tuning for cognitive state decoding, neurofeedback systems, and human-computer interfaces — without requiring extensive per-subject pretraining.
But here's the thing that gets underappreciated: the most powerful version of this technology is not a model built for one industry. It's a foundation model — trained on high-quality, diverse EEG data across contexts, subjects, tasks, and environments — that then becomes a platform. The same way GPT didn't need to be retrained from scratch for every application, a well-built large brain model shouldn't need to be rebuilt for every vertical. A construction company shouldn't have to solve the neuroscience. A surgical robotics company shouldn't have to solve it either. They should be able to build on top of a shared, generalizable cognitive state layer — and focus on the application.
What's missing today isn't just integration within one industry. It's the foundational model layer itself — trained at sufficient scale and diversity that entrepreneurs across healthcare, heavy industry, defense, and manufacturing can build on top of it without each reinventing the science. The neurotechnology market is predicted to reach $21 billion by 2026, with significant enterprise applications emerging across workplace safety and cognitive performance monitoring. That market won't be won by vertical-specific models. It will be won by whoever builds the brain equivalent of a foundation model that others can fine-tune.
EMOTIV's work on large brain models trained on decades of EEG data is exactly this direction. The hardware exists. The data is accumulating. The models are maturing. What the ecosystem now needs is entrepreneurs who treat cognitive state as an input layer — the same way they treat GPS, computer vision, or voice recognition today.
If EEG-based models can decode intent and state reliably enough to replace a keyboard interface — what does that mean for the operator running a 50-ton machine from a remote station?
Research has already demonstrated EEG-based BCI systems that translate a construction worker's imagined body movements into robotic commands with over 90% accuracy. EEG sensors designed specifically for crane operators can detect mental fatigue in real time, providing early warning before reaction speed degrades and accident risk climbs.
The capability exists. What's missing is integration — and intent.
Right now, teleoperation systems are designed around machine performance. The feedback loop is: machine does something → operator corrects → machine adjusts. It's reactive and one-directional.
Add cognitive state awareness and the loop changes fundamentally. Now you have: operator enters high-fatigue state → system detects it → system intervenes. That could mean slowing autonomous assist, flagging a supervisor, rotating the operator, or in more advanced configurations, engaging a co-pilot mode. Brain-robot interaction systems have advanced significantly over the past decade, with EEG-based control now enabling both active and passive modes of machine operation across a range of robotic systems.
This is not science fiction. It's what aviation did decades ago with crew resource management and physiological monitoring. Construction has never applied equivalent rigor to the human in the cockpit.
The construction industry needs approximately 499,000 net new workers in 2026, with 41% of the current workforce projected to retire by 2031. The skilled labor shortage costs the home building sector alone $10.8 billion per year, with 45% of firms reporting project delays directly caused by worker shortages.
As the pool of experienced operators shrinks, the ones remaining are working longer shifts, running more machines, and making higher-stakes decisions under greater cognitive load. That's exactly the environment where fatigue-driven errors compound — and where the stakes of getting it wrong are highest.
A less experienced remote operator managing two machines from a control center is not the same as a veteran in the cab. Cognitive state monitoring bridges part of that gap — not by replacing skill, but by making invisible degradation visible before it becomes an incident. And critically — because there is still a human accountable in the loop — the insurance and liability frameworks that have blocked full autonomy adoption don't apply here. This is a layer of intelligence added to a human-operated system. That distinction matters enormously for real-world deployment.
The real infrastructure gap isn't in any single vertical. It's upstream.
What the industry needs is a large brain model trained on high-quality, diverse EEG data — not scoped to construction, or healthcare, or defense — but broad enough that it learns generalizable representations of human cognitive and emotional states across contexts. Once that foundation exists, the application layer opens up. A startup building teleoperation software for mining equipment can plug in cognitive state as an input. A surgical robotics company can use the same model to monitor surgeon focus during a procedure. A logistics operator can use it to flag fatigue in remote forklift operators. None of them need to solve the underlying neuroscience.
This is the infrastructure play. And it mirrors exactly how every other major AI platform has scaled — not by building narrow models for each use case, but by building generalizable foundation layers that others build on top of.
The technology stack to unlock this already exists in pieces: industrial-grade EEG headsets, foundation models that generalize across subjects without weeks of individual calibration, and teleoperation platforms with open APIs. The missing piece is the commitment to building — and openly enabling — a shared cognitive state layer that the next generation of human-machine systems can be built on.
The construction and heavy equipment industry is a $200 billion market building telematics infrastructure that knows everything about its machines. The next frontier is knowing something about the people running them. And the entrepreneurs who build that layer won't just transform one industry. They'll transform every industry where humans and machines share control.
1. KU Leuven / VSC — ST-EEGFormer, ICLR 2026 & NeurIPS 2025 EEG Challenge — vscentrum.be
2. Chen et al. — Large Cognition Model: Towards Pretrained EEG Foundation Model — arXiv:2502.17464
3. Cheng et al. — Brain-computer interface for hands-free teleoperation of construction robots — ScienceDirect (2021)
4. Mental Fatigue Detection of Crane Operators via EEG — PMC/NCBI (2025)
5. Global Market Insights — Heavy Equipment Telematics Market — gminsights.com (2026)
6. Associated Builders and Contractors — 2026 Construction Workforce Report — abc.org
7. Home Builders Institute / NAHB — Construction Labor Market Report Fall 2025 — nahb.org
8. Nik Aznan et al. — Mind Meets Robots: EEG-Based Brain-Robot Interaction Systems — Taylor & Francis (2025)
9. The Conversation — Brain monitoring may be the future of work — theconversation.com (2026)
10. Research Gate — Fatigue and workplace accidents in construction (2025)
11. Growth Market Reports — Autonomous Excavator Liability Insurance Market (2025)
