Building a robot is the easy part. Teaching it not to panic on a chaotic construction site is where things get interesting.
The real hurdle in construction robotics is not building hardware. It is giving machines the intelligence to think, adapt, and operate safely alongside human workers. To find out how the industry is solving this, we sat down with Sally Walkom, Co-Founder and Chief Partnerships Officer at PRENTEN.AI.
With nearly 25 years of deep experience spanning property and infrastructure transactions and innovation, Sally has made a bold leap into robotics and physical AI, focusing entirely on making the technology operationally viable for construction and other high risk industries.
We asked Sally three questions to give you an insight on her mission.
What does it take to give a construction robot a mind?
Construction sites are unpredictable environments, with people, machinery, materials and processes constantly changing. What are the key ingredients required to enable a robot to understand its environment, make decisions and adapt to what is happening around it rather than simply follow a programmed sequence?
SALLY WALKOM:
There are three layers to this, and they build on each other.
The first is a digital twin combined with object recognition. This is classical machine learning that gives the robot a live picture of what is on site. A static operations and maintenance environment and an active construction site are completely different. One changes slowly, while the other changes hour to hour. The robot is therefore equipped to tell the difference.
The second layer is a physical AI model that gives the robot an understanding of physics. That is the difference between a robot that can label what it sees and one that can reason about what is likely to happen next.
The third layer is autonomous navigation. This involves letting the robot move through that environment and make its own calls. Right now most of our navigation is route programmed, but full autonomy is where we are heading once the first two layers are built.
Can robots learn from the way humans work?
There is growing interest in using closed circuit television, site video and other forms of observational data to capture how people actually move, work and interact with equipment. Could the enormous amount of human activity already being recorded on construction sites become a training resource for robots? And what would it take to turn that data into useful robotic intelligence?
SALLY WALKOM:
At this stage, human activity is more noise than signal for us. When we are building the digital twin of a site, a busy construction environment is actually harder to map cleanly than a quiet one. So the near term priority is getting a reliable read on the physical environment, and we actually mask any people captured to get that clear picture. There is also a trust question. In construction, people are understandably wary of cameras and sensors that could be seen as surveillance rather than tools that help them. We are not interested in building a system that watches workers to monitor them.
Rather, we are interested in human robot teaming. This means using the skills and judgment people have to direct what robots do, and using robots to fill gaps where sites are short staffed, short on time, or potentially unsafe. With humanoids down the track, human demonstration is a useful way to teach a robot manipulation tasks. For quadrupeds, whose job is inspection and data capture rather than physical manipulation, that training pathway does not apply in the same way.
Where does construction robotics go from here?
As robots become increasingly capable of perceiving, learning and acting in the physical world, how do you see their role developing over the next five to ten years? Will we see robots mainly taking on specific repetitive or hazardous tasks, or could they ultimately become autonomous members of project teams capable of adapting to much broader construction activities?
SALLY WALKOM:
The future potential is not just with repetitive and hazardous tasks. Robots will increasingly take on tasks people simply do not want to do, and that captures more than just the dangerous or the boring. However, I do not think this future happens by default. We have to actively shape it.
The future we are building toward is not about cutting jobs. It represents a shift in what being on the tools actually means. Someone who is doing hands on inspection or data capture today becomes the person supervising a team of robots doing that work, and applying their judgment to what the robots find. That is a much more valuable use of their experience.
And this matters for construction because we have a capacity problem. Australia needs more homes, more public transport, more community facilities, and more energy infrastructure. Crucially, we do not have enough people to build all of it at the pace required. If robots let us do more with the workforce we already have, that is not a threat to jobs. It is how we actually get the pipeline built.
The roadmap for autonomous infrastructure is being drawn right now, and the minds turning theory into reality are already moving fast. To meet Sally and other leaders shaping the future of construction technology, check out the Construction Technology Leaders Summit.
