THE ROBOTICS DEPLOYMENT LAYER BUILDING NOW
Closing the loop
between robots and
the physical world.
Zonta Labs is building drones and software for teams that deploy robots. The drones are designed to watch each run and rebuild it in 3D, so AI agents get the physical context to see what went wrong and improve the next run.
RECORDED SIMULATION / DRONES MAP A LAB TRACK IN 3D
Eyes on the robot.
A compact drone, designed to watch robots work.
Perceive, interpret, adapt, act, repeat.
Robot learning still has a person in the loop. When a run fails, someone has to work out why before the robot can be retrained and sent back out.
Zonta is building a real-time perception and interpretation layer to replace the manual loop in robot deployment. It is a product for teams to run with their own robots, not a service that deploys robots for hire. Every run should produce feedback the next run can use.
Perceive
DRONES
While the robot works, drones watch it, the task in front of it and the space around it.
Interpret
3D SCENE + AGENT
What the drones saw becomes a 3D scene of the run. A frontier AI agent reads it to understand what the robot did, what came of it and where it failed.
Adapt
POLICY + SIMULATION
The agent turns that understanding into changes: an adapted policy, an updated simulation, or new training scenarios aimed at the failure.
Act
ROBOT
The robot redeploys with those changes and tries the task again, with the drones still watching.
Repeat
RSI RL
Each run feeds the next, with less need for an engineer to step in after every failure.
RSI RL, Recursive Self-Improvement Reinforcement Learning, is our name for this loop once it runs without a person in the middle.
What works today, and what we are building.
Zonta Labs is an early-stage company. Everything listed as working today is on this site. The rest is in development.
WORKING TODAY
IN DEVELOPMENT
- Real-time perception
- Physical context for AI agents
- Adaptation
- Developer portal
WHERE IT GOES
- RSI RL for robotics
The people building it.
Zonta Labs is three co-founders, with backgrounds in electrical engineering, computer science and physics.
Andrew Wang
- Role
- Co-founder
- School
- UC Berkeley
- Field
- Electrical Engineering and Computer Sciences (EECS)
Abhirama Sonny
- Role
- Co-founder
- School
- UT Austin
- Field
- Computer Science and Physics
Daniel Zhao
- Role
- Co-founder
- School
- Harvard
- Field
- Computer Science