Gritt AI

Real job sites... but with thousands of simulated variations.

Gritt turns field data into synthetic worlds to train our construction superintelligence, then feeds every deployment back into the loop.

Our CTO Vishal Dugar explains how:

A worker kneeling beside a solar panel is doing ordinary construction work.

But to a robot, successfully identifying that worker may present an edge case: the worker is low to the ground, partly hidden by equipment, and wearing heavy PPE.

The problem in identifying the worker starts before model training.

Annotation pipelines often reject images when they can’t confidently reconstruct a worker’s full pose. That may improve the dataset's apparent quality, while removing the difficult examples a safety-focused system most needs to learn from.

At Gritt, we call this the selection-scarcity confound.

After auditing four months of construction-site data, we found that low poses accounted for just 2.2% of the retained person instances. Further analysis suggested that this scarcity was partly created by the filtering process itself.

So we separated person detection from pose-label quality, then brought the missing postures into simulation. Adding simulated workers who were kneeling, squatting, or deeply bent over improved detection recall on real site footage for those poses by 1.6 percent and low pose detection by 5 percent.

In physical AI, rare examples aren’t necessarily unimportant... in fact, they may be the *most* important when it comes to safety.

Read the full post from Gritt Perception Lead, @r15hav, with a link to the paper on arxiv:
gritt.ai/news/the-safet…

Gritt is creating a construction superintelligence that can build in decades what would take centuries today.

To do that, we've built a generalizable AI backbone that learns representations of the construction site and applies them to any dextrous construction task. This AI is embodiment-agnostic, so we can bring the right machine-and-robotics combination to the site and task that works best, as quickly as possible.

We're already deploying across the country, starting in solar and rapidly expanding to new verticals.

Large-scale infrastructure construction sites push physical and spatial AI to their limits.

In Gritt’s first technical article, CTO and co-founder Vishal Dugar explains how field data, simulation, and our foundation model build practical intelligence that generalizes across tasks, environments, and embodiments.

gritt.ai/news/how-gritt…

Going to Pittsburgh is a bit of a homecoming for our founding team of CMU Robotics Institute alums.

Join Gritt at #IROS2026 on Monday, September 28th to meet the team, connect with robotics peers, and hear about the traction we’ve gained in deploying a physical AI intelligence layer in the field.

RSVP: luma.com/w1zz14gu