Robots can be programmed to perform various tasks akin to packing boxes and even performing surgery. However, each individual movement or task requires its own specific training process, making it difficult for robots to adapt to real-world scenarios.
Mbodi desires to make it easier and faster to train robots using AI agents. The company will showcase this technology as one of the top 20 Startup Battlefield finalists at TechCrunch Disrupt 2025.
New York-based Mbodi has built a cloud-to-edge system, a hybrid computing system using each cloud and on-premises computing designed to integrate with existing robotics technology stacks. The software relies on multiple AI agents that communicate with each other to assemble needed information to assist the robot learn a task faster.
Once deployed, Mbodi will collect data and learn from real-world use cases.
Xavier Chi, co-founder and CEO of Mbodi, told TechCrunch that users prompt the software using natural language, and Mbodi breaks down the request into smaller subtasks. Mbodi’s group of agents essentially divides and carries out the task of gathering the information needed to quickly train the robot on the fly.
“The difficult thing about the physical world is the endless possibilities,” Chi said. “Any time you can invent something completely new, you don’t have any data, that’s a problem in the physical world. We always have to have a system where you can organize different models or have someone tweak the robot and tell it to do certain things in a certain way.”
Chi said he and co-founder Sebastian Peralta got here up with the idea for the company while working as engineers at Google. Although they weren’t working on robotics, the two realized that advances in artificial intelligence were heading into the physical world, and despite advances in physical AI, there was still no good approach to train robots quickly.
Techcrunch event
San Francisco
|
October 27-29, 2025
Many firms, akin to Skild AI and FieldAI, need to speed up training robots by building large AI models around the world with enough real-world data to assist them adapt to recent environments. Chi said philosophy just doesn’t work given the world is continually changing.
Mbodi was launched in 2024 with a focus on picking and packaging. Last yr, the company won the ABB Robotics AI startup competition, because of which it established cooperation with the Swiss robotics organization, acquired by SoftBank in October for $5.4 billion.
The company is currently working with a Fortune 100 consumer and product company on a proof of concept.
“The CPG customer is a company that employs a lot of people and packages different branded products on a tray or shelf, the problem is that the products change every day,” Chi said. “For this reason, you can’t put robots there. It’s simply impossible to reprogram these robots because there are still so many people doing this work.”
Mbodi hopes to start a wider rollout of its software in 2026.
“We want to build something that works and that is applicable,” Chi said. “We are not a research laboratory; we do not want to be a research laboratory in this respect. We want to put something into production that works reliably.”