Autonomous capture will quickly become table stakes. The important question will be which platform can orchestrate robots and agents together in a unified, automated and intelligent way to create the most value.

The Biggest Impact of Autonomous Capture and AI
The Biggest Impact of Autonomous Capture and AI

Q&A with Nicholas Pilkington, CTO and co-founder, | DroneDeploy

Tell us about yourself and what led you to found DroneDeploy.

I’ve been an engineer my whole life and like building systems that touch the physical world. I grew up in South Africa and have known my co-founders, Mike Winn and Jono Millin, since high school.

The concept really took shape around Kruger National Park, which is one of Africa’s largest game reserves. There are roughly 650 rangers trying to protect more than 7,500 square miles of wilderness, so there is simply no way to patrol every corner on the ground. Mike had been working with park rangers on anti-poaching operations and the idea was straightforward – launch a drone with a thermal payload, fly systematic transects over the bush at night and spot poachers before they reached the rhinos. In practice you ended up with Windows laptops on the hood of a truck, large ground stations, command-line tools and almost no autonomy.

DroneDeploy was our attempt to change that. You plan a mission on your phone, let the autopilot and our photogrammetry engine do the work and a usable map or 3D model appears in the browser. That is still the core idea: make the real world observable and computable from anywhere.

 

DroneDeploy started in aerial reality capture and is now working much more deeply in robotics and AI. How do you describe that evolution, and what are you ultimately trying to enable on a typical construction site today?

The first phase was about taking drones from RC toys with cameras to reliable survey instruments. We built a cloud flight stack and a photogrammetry engine so anyone could get orthomosaics and 3D models out of a drone or docked system without being a GIS specialist. On a large site today, weekly dock flights and overlays are simply how you keep track of earthworks and structure.

But the challenge with manual capture is that humans, your most valuable assets, still have to do that work. The next phase is what you are seeing now, with autonomy and AI agents layered on top of that visual data. Once air and ground capture are autonomous, the bottleneck is no longer getting pixels, it is people having to interpret those pixels. So we are wiring drones, ground robots and 360 capture into a common spatial model and letting Progress AI and Safety AI reason over it using LLMs and vision models. On a healthy project the goal is that the site documents itself and each morning the team sees a current, machine-generated picture of what changed, what is off-plan and where risk may be creeping in.

 

When you talk about embodied AI on the ground side, what does that mean in practice, and why is this moment different from earlier experiments with robots on jobsites?

When we say embodied AI, we mean taking an AI agent and giving it a body, a sensor stack and a very tight mission to profile. In our world that is usually a quadruped or compact UGV (unmanned ground vehicle) with 360° cameras, depth sensors and an onboard computer module running SLAM (simultaneous localization and mapping) for navigation as well as sensor management. Its job is boring but critical – walk the same routes every day through corridors and rooms that a drone cannot reach and feed high-fidelity data into the visual twin.

Right now the full stack is finally ready for production work. Early robots were expensive, fragile and needed a dedicated handler, so they were used for marketing more than true autonomous capture. New platforms are affordable, more reliable and capable of obstacle avoidance, navigating busy floor plates and docking safely to recharge. Connectivity has caught up as well, with Starlink, mesh Wi-Fi and better edge-to-cloud pipelines that let us blend autonomy, teleoperation and remote monitoring. Because Progress AI and Safety AI are trained on large construction datasets, every walk turns into useful progress and safety signals instead of another disk full of video.

 

If we follow a construction project from the end of one day to the start of the next, how do the different pieces – drones, ground robots and AI – work together behind the scenes?

The rhythm looks fairly consistent across good deployments. Docked drones are unobtrusive and can fly multiple times a day. Once in the morning as the sun comes up to capture the “as is” version of the site for the day, and once again towards the end of the day after work has stopped. The aerial drones fly their pre-planned missions, updating the exterior model and producing fresh orthomosaics and point clouds to be used for earthworks quantification, logistics planning and progress tracking. As the trades wind down in the afternoon, the ground robots undock and run their interior routes, weaving around materials, scanning corridors and capturing 360 imagery and LiDAR across every area on site – a comprehensive visual record of the site for the day. All of that feeds into the same photogrammetry and spatial indexing pipeline so you have an updated visual twin of the job within hours.

The moment data has gone through 3D reconstruction or static imagery hits the cloud, the AI agents go to work. Progress AI compares what it sees against the schedule using a mix of design data, computer vision and LLM-based reasoning. Safety AI scans for risk patterns and compliance gaps. By the time the superintendent logs in, they are not hunting through folders. They have a concise, machine-generated view of where work advanced, where it may be stalled and what needs attention. Robots and drones provide the capture layer and the agents provide the interpretation and translation into business value.

 

Where are you seeing autonomous capture and AI make the biggest impact so far, and how are customers measuring that impact?

The strongest impact right now shows up on projects where guessing is very expensive. Hyperscale data centers, pharma and life-science facilities, energy and renewables and large industrial programs all fall into that category. Those jobs have dense MEP (mechanical, electrical and plumbing engineering), tight tolerances and liquidated damages that turn schedule slips into serious dollars.

Customers usually lean on a mix of financial and operational metrics. One lens is hard cost and rework; continuous aerial and ground capture lets teams catch design-versus-field mismatches early, and we routinely see savings on the order of $10,000 per $1 million of construction through fewer rip-and-replace events, better quality control and less travel to site. Another lens is risk and claims. An unbiased, time-stamped visual twin has been shown to drive a 40% reduction in claim size and value in research by our partner Shepherd, with up to a 25% premium discount when DroneDeploy is part of the workflow. A third is schedule and margin, where Progress AI and Safety AI help teams protect the critical path by surfacing issues when they are still cheap to fix.

 

There are a lot of tools and vendors in this space now. From your perspective as CTO, what really differentiates a true platform approach from point solutions, and where do you think the industry will be a few years from now?

At DroneDeploy, we focus on three pillars that customers tell us matter most when they choose a robotics and AI platform for the long term.

The first pillar is Unified. Many 360 capture solutions have bolted on entry-level drone capabilities, but they do not have the geospatial and positioning foundation you need to align data with design intent or repeat captures over time. The first time you show a field superintendent inaccurate or misaligned data, you lose their trust. Accuracy is the foundation of that trust and a big reason DroneDeploy is selected when the stakes are high.

Once you can trust the data, the real cost of data collection sits with humans walking sites. That leads to the second pillar, Automated. Here we focus on autonomous robotic capture from both the air and the ground. We acquired a robotics company, Rocos, in 2021 and have been developing our robotics platform to plan and execute autonomous missions for the last five years. Because we now have mission-planning software for aerial docked drones and ground robots, we can deliver a fully autonomous inspection of an entire site, which is something the point solutions in our space are not approaching today.

The third pillar is Intelligent. Autonomous capture creates a new problem, which is a mountain of imagery that no human could ever review end to end. This is where we layer in AI agents such as Progress AI, Safety AI and Inspection AI. Those agents call into LLMs and vision models with full context about geometry, time and location. About a year ago we made a deliberate decision to spin up a new codebase and build these agents from the ground up, using modern AI to deliver insights 100 times faster and at a fraction of the cost of legacy approaches. Results now arrive in minutes instead of days, and that time difference often decides whether you catch an issue or miss it.

Autonomous capture will quickly become table stakes. The important question will be which platform can orchestrate robots and agents together in a unified, automated and intelligent way to create the most value.

 

As this technology matures, how do you see the roles of supers, project teams, and owners evolving alongside more automation on their sites?

I do not expect a future where a robot replaces a superintendent. What I expect is a future where the superintendent has much better instrumentation. Instead of walking the site with a phone just to prove something happened, they can rely on a continuously updated visual twin and AI-generated insight to drive coordination, sequencing and safety conversations. The emphasis shifts from gathering evidence to making higher-quality decisions in less time, compressing the schedule.

For project engineers, VDC and schedulers, autonomous capture and AI effectively turn the site into a streaming data source. That changes how schedules, RFIs, change orders and delay analyses are managed because teams can go back to ground truth rather than fuzzy memory. Owners and insurers see a similar step change as they move from episodic reporting to continuous, objective visibility across a portfolio, which in turn influences how they think about risk and capital deployment. In all of these cases, automation keeps people in the loop but pushes their attention up a level, away from manual documentation and toward system-level thinking about how to build and operate better.

 

Nicholas Pilkington is the co-founder and Chief Technical Officer of DroneDeploy, the robotics and AI platform used on over 3 million sites worldwide. As the only platform that combines drones, robots, 360 cameras and AI agents, DroneDeploy captures and organizes site conditions into a single, time-stamped record that teams can trust in the field and the office.

Nicholas leads technical strategy at DroneDeploy and is responsible for innovation and research leadership. He completed his PhD in Machine Learning at the University of Cambridge.

 

The content & opinions in this article are the author’s and do not necessarily represent the views of RoboticsTomorrow

Featured Product

DESTACO - Revolutionizing Industrial Automation

DESTACO - Revolutionizing Industrial Automation

Looking for a reliable solution to enhance your automation process? Look no further than the DESTACO Robohand Grippers. These grippers are designed for the modern world of robotics, offering unparalleled performance and precision. Whether you need to grip fragile items, irregularly shaped objects, or heavy-duty components, the DESTACO Robohand Grippers have got you covered. Their modular design allows for quick and easy customization, ensuring a perfect fit for your application.