Trace · New York
OtherResearchFullTimeposted 6d ago
About Trace
Trace is building the data infrastructure that lets robots learn from the real world
AI is moving into the physical world. It has the potential to transform how work gets done in the real world, from robotics to embodied systems that can see, move, and interact with their environment. But today, progress is constrained by a fundamental limitation: there is no scalable way to collect high-quality, real-world training data. Frontier robotics models are trained on orders of magnitude less data than language models because there is no equivalent of an "internet of robotics data."
Trace exists to change that.
We build the infrastructure that makes it possible to capture and transform real-world data like humans performing physical work, and turn it into training data for robots and other intelligent systems that operate outside the browser and in the physical world.
If we succeed, we will meaningfully accelerate the development of physical AI and expand what these systems can safely and reliably do in the world. If you want to be an early hire at a company helping define how robots learn to work, keep reading.
Why Trace
A world-changing problem: Physical AI will reshape entire industries, but it cannot scale without real-world data. Trace is addressing one of the core constraints holding the field back.
Early with real traction: Demand for real-world data is exploding as we start to see the impact of intelligent physical systems.
Experienced, tight-knit team: Ex-founders and operators with a track record of building and scaling together.
Real ownership: This is early. Your research will materially shape what the company trains, how it evaluates progress, and where it invests next.
A research team, not a publication pipeline: We care about work that makes our systems better. You'll be close to real data and real deployment.
The Role
We're hiring a researcher to join Trace and work on the core problems in robot learning: what data quality means in the paradigm of robot learning. You’ll work on how to structure real-world data for training, how to train and evaluate models on it, and how to close the loop between what we capture and what makes physical AI systems better.
This role reports to our Chief Scientist. You'll be expected to do frontier research on data quality, running experiments that will push forward our understanding of how data moves model performance. We're looking for people motivated by seeing their research move a system forward, with strong technical opinions, weakly held.
What You’ll Do:
Design and run experiments on training robotics, vision, and embodied AI models using Trace's real-world data.
Investigate how data quality, structure, and diversity affect downstream model performance, and turn those findings into concrete recommendations for what we capture.
Build and iterate on evaluation methodology so we know whether a model, or a dataset, is actually getting better.
Stay close to the literature on robot learning and translate what's relevant into work we can actually ship.
Work directly with the engineering and data teams so research findings turn into pipeline and product changes.
Who You Are:
Currently completing, or recently completed, a Master's or PhD program with a research focus on robot learning, robotics, or training foundation/AI models.
Exceptional undergraduates from top-tier robotics programs will also be considered.
No industry experience required. You can join us straight out of your program.
If you're coming from industry rather than a research program, your background should be a researcher role at a robotics company or an equally reputable research lab. General industry engineering experience without a research track record isn't a fit for this role.
Motivated by seeing research translate into working systems
Optimistic and serious about where robotics and AI are headed.
Comfortable being early in your career but still forming strong, defensible technical opinions.
Bonus:
Publications or research work specifically in robot learning or training models for physical/embodied systems
Experience from a leading robotics research program or lab
Hands-on exposure to real robotic hardware or real-world (versus purely simulated) data