Design
Help bring real, skilled work into AI research.
We’re inviting companies with machines and experienced operators to help create public datasets: recordings that other engineers can use to study real work. For selected partnerships, Andes Path contributes engineering at no charge. Together, we collect useful data, run an early AI experiment, and publish and promote the results.
A project can start with one machine, an experienced operator, and a focused collection of work. You don’t need an AI department, years of recordings, or a finished research proposal.
You get direct access to our engineering team, firsthand experience with the technology, and a real contribution your company can put its name on. The research community gets something it can build on.
AI has a lot to learn from your world
An experienced operator makes difficult work look routine. They adjust as the ground changes, a load shifts, or the material behaves differently than expected. What looks like “moving a bucket” from a distance can draw on years of experience.
That complexity is easy to lose when a job becomes an abstract research problem. A tidy demonstration doesn’t necessarily capture the judgment and small corrections involved in getting actual work done. We want more of that reality in AI research, and the public starting points can be surprisingly small.
One excavator dataset from Flywheel and its collaborators contains just 25 hours of operation, connecting camera recordings with joystick movements. Link: Hugging Face
For a different task, PSIORI released 175 sensor scans to help computers recognize parts of a working crane. The team specifically identified a shortage of public data for that kind of equipment. Link:GitHub
These are individual releases, not totals for their industries. A meaningful contribution is often a focused, small collection. What your team considers an ordinary day could give researchers something they would otherwise struggle to study, and their research might one day make your work easier, faster, or more effective.
In the example below, our engineers used about 13 minutes of recorded excavator work to adapt π0.5, an AI model developed by Physical Intelligence to control robots. Rather than starting from scratch, we built on a model already trained on other robotic tasks and explored what it could learn from this small sample of operator demonstrations. The video visualizes the excavator movements that the model predicts. This is an early research experiment, not a self-driving excavator. It illustrates the kind of project we want to make possible: turning a focused sample of skilled work into something researchers can build on.
Work with the engineers behind physical AI
Andes Path is a specialist robotics software engineering firm. We build the software that powers and supports physical AI systems. If you operate machines in the real world, that either includes you, or it will include you soon.
The data pipelines, remote-operation systems, simulation, fleet management, and developer tools are what make robotics work in real life, reliably. More than 1,000 robots run on systems we’ve built. This type of software is our core business.
Through these partnerships, we’re opening our expertise to companies that know the machines and the work but don’t have a robotics engineering team of their own. Your people bring the practical knowledge. Our engineers help turn it into a useful research contribution. Over time, that couuld lead to entire step changes in your industry.
Start learning now (so you won't have to catch up)
You don’t have to wait for a finished autonomous machine to start understanding the technology behind it.
Working alongside our engineers gives your team a chance to see what current AI can do, where it struggles, and what makes your work difficult to capture. You can ask questions, challenge assumptions, and develop your own informed view rather than relying on someone else’s demonstration.
Your company also gets something concrete to share. We publish and promote the work together, crediting your team’s contribution and telling the story of the project: what we investigated, what we built, and what we learned.
That gives you more than an announcement about “embracing AI.” It gives you a real project to discuss with customers, employees, and industry peers, and experience to draw on as you consider what comes next.
You bring the access. We help handle the data.
Existing recordings are welcome, but they aren’t required. We can help plan a manageable collection around your equipment, people, and operations.
We want to connect what the operator sees, how they use the controls, how the machine responds, and what happens next. Our engineers help capture, organize, and prepare those recordings so they can support an experiment and be useful to other researchers.
Sensitive information does not have to rule out a conversation. We can help clean and sanitize the data so privacy is preserved. For example, blurring faces and identifying details, removing customer or location information, and excluding restricted footage are all part of any project.
We'll agree upfront on how data will be collected, handled, reviewed, and shared. The public release is the agreed, reviewed portion, not everything we capture. We will always work within your constraints.
You don’t need to arrive with perfectly prepared data, and you don’t need to open your entire operation to the public.
These are early sparks. Sharing them helps the research grow.
One dataset will not make a jobsite autonomous or capture everything a skilled operator knows.
We’ll adapt an existing AI model to your focused task and measure what it can learn. The outcome may show promise, expose a limitation, or reveal something else we need to record. The commitment is a useful prototype or evaluation, not production-ready autonomy. But it's a great first step.
Publishing the data and findings lets the work continue beyond our first experiment. Another team can try a different approach. Others can compare results, contribute recordings from different conditions, or investigate a question we haven’t considered.
Instead of every team starting by finding a machine and arranging its own collection, they have somewhere to begin.
That is the opportunity: together, we'll help make the next round of research possible.
Tell us about the work
We’re interested in construction, agriculture, forestry, subsea work, fabrication, and other settings where people operate machines to do skilled physical work.
Tell us what equipment you operate, what your people do with it, and what access you could provide. We’ll work together to identify a worthwhile, manageable project.
Andes Path will contribute a defined allocation of engineering time at no charge. Your company provides equipment access, operator time, collection support, and permission for the agreed public release and joint promotion. Before starting, we settle the scope, responsibilities, publication plan, and any equipment or compute costs. There is no requirement to purchase further engineering work.
If that sounds interesting, email hello@andespath.com to start the conversation. We'll get back to you quickly.
Your people know the work. Our engineers can help bring it into AI research. Together, let’s give the community something real to build on.
