World’s First Fully Automated Carpet-Cleaning Robot Revolutionizes “Hands-off” CleaningWorld’s First Fully Automated Carpet-Cleaning Robot Revolutionizes “Hands-off” CleaningWorld’s First Fully Automated Carpet-Cleaning Robot Revolutionizes “Hands-off” Cleaning AnaTech Maz Technology Magazine
 
  1. Home
  2. Future Technologies
  3. Robotics

Mbodi to Demonstrate AI-Powered Robot Training at TechCrunch Disrupt 2025

Priyadharshini S November 03, 2025 | 3:09 PM Technology

Robots can perform a wide range of tasks, from packing boxes to assisting in surgery. However, each specific movement or task usually requires its own training process, making it challenging for robots to adapt to real-world scenarios.

Figure 1. AI Agents to Train Robots: Mbodi at TechCrunch Disrupt 2025.

Mbodi aims to simplify and accelerate robot training using AI agents. The company will showcase its technology as one of the Top 20 Startup Battlefield finalists at TechCrunch Disrupt 2025. Figure 1 shows AI Agents to Train Robots: Mbodi at TechCrunch Disrupt 2025.

Based in New York, Mbodi has developed a cloud-to-edge system—a hybrid computing setup that leverages both cloud and local processing—designed to integrate seamlessly with existing robotic tech stacks. The software uses multiple AI agents that communicate with one another to gather information, enabling robots to learn tasks more quickly and efficiently.

Xavier Chi, co-founder of Mbodi, told TechCrunch that users interact with the software through natural language prompts. Mbodi then breaks these prompts into smaller subtasks. Its cluster of AI agents essentially “divide and conquer,” gathering the information needed to train the robot on the requested task quickly.

“The tricky thing with the physical world is its infinite possibilities,” Chi said. “Every time you try something completely new, there’s no prior data—that’s a real challenge. You need a system that can orchestrate different models or allow someone to correct a robot and guide it to perform tasks in specific ways.”

Chi explained that he and co-founder Sebastian Peralta conceived the idea while working as engineers at Google. Although they weren’t focused on robotics at the time, they recognized that AI was increasingly moving into the physical world. Despite growing advances in physical AI, there still wasn’t an effective way to train robots quickly.

Many companies, such as Skild AI and FieldAI, are trying to speed up robot training by building large “world AI” models with massive amounts of real-world data to help robots adapt to new environments. Chi, however, said that approach falls short because the real world is constantly changing.

Mbodi, which launched in 2024 with a focus on picking and packaging, won an ABB Robotics AI startup competition last year. This victory led to a partnership with the Swiss robotics organization acquired by SoftBank for $5.4 billion in October.

“For the CPG customer, they have many employees packing different products into trays or shelves, but the setup changes every day,” Chi said. “Because of that, it’s impossible to rely on robots. Reprogramming them isn’t feasible, so humans are still doing much of the work.”

“We want to build something that actually works and can be deployed,” Chi said. “We aren’t a research lab; our goal is to put a reliable system into production.”

Source: TC

Cite this article:

Priyadharshini S (2025), Mbodi to Demonstrate AI-Powered Robot Training at TechCrunch Disrupt 2025, AnaTechMaz, pp.292

Recent Post

Blog Archive

Popular Lnks