Mbodi is working with partners such as ABB Robotics. Source: Mbodi AI
Manufacturers are facing worsening labor shortages, but programming conventional robots to respond to changes in production can take weeks. As a result, billions of dollars in high-mix, low-volume manufacturing remains out of reach for automation, according to Mbodi AI.
The New York-based company today announced that it has been selected to join the Y Combinator accelerator’s X25 batch.
“We’re building an embodied AI platform that lets anyone teach robots new skills — just by talking to them. No code. No engineers. No week-long reprograms,” stated Mbodi AI co-founders Xavier (Tianhao) Chi and Sebastian Peralta. “Mbodi turns natural language and quick demos into precise, reliable robot actions that run in production in minutes, and adapt as often as you need.”
Chi and Peralta, who helped run Google Public DNA (8.8.8.8) at Google, founded Mbodi AI last year. In December, ABB named the New York-based company as a winner of its 2024 ABB Robotics AI Startup Challenge.
In addition, Mbodi has been recognized with a 2025 RBR50 Robotics Innovation Award. Chi spoke in a panel on generative AI’s impact on robotics at the Robotics Summit & Expo earlier this month.
Xavier Chi and Sebastian Peralta were engineers at Google before co-founding Mbodi. Source: Mbodi AI
Automation needs flexibility for adoption
Manual labor drives half of gross domestic product (GDP), or $40 trillion, but 70% of factories worldwide face labor shortages, according to Mbodi AI. Why hasn’t automation spread more quickly?
“It’s interesting — when I got into the field, I found that the world is a lot less automated than I thought,” Chi replied to The Robot Report. “For instance, people think packing is mostly automated, but a lot is still done by humans.”
Reprogramming robots requires on-site specialists and time, he said. Manufacturers can’t keep up with ever-shifting production needs and growing demand, so they need an easier way to instruct robots.
“Most tasks require some level of flexibility, and automation usually involves a long development cycle,” said Chi. “On the manufacturing floor, people don’t have the expertise to run these robots. Collaborative robots were originally designed to be sold directly, but end customers may still have a hard time with integration and cost, which are barriers to mass deployment.”
Now accepting session submissions!
Mbodi AI helps robots learn tasks
Mbodi said its cloud-to-edge system combines generative AI, agent orchestration, and symbolic reasoning to turn high-level instructions into precise, safe, and adaptive robot actions. The startup claimed that its “universal agent framework can run across any robotic hardware, coordinating perception, reasoning, planning, and control in real time.”
The company sees agentic AI as invisible to the user, who just tells the robot what to do, said Chi. One agent might get data from the world, while another might break down tasks or spawn new agents.
“We like the scene in Iron Man where the robot just follows instructions and passes Tony Stark tools,” Chi recalled. “It was fiction before, but it’s coming to life now, thanks to generative AI.”
“Currently, most robots have to be programmed to do a specific thing, so you need integrators,” he added. “That has been the case since these robots were invented decades ago. But now, you can repurpose a robot more easily today and tell it what to do.”
Mbodi AI’s interface abstracts some of the control layer, bringing it together with vision for object detection and path planning, for a more generalized intelligence, explained Chi.
Chi also noted that when one robot learns a task, it can share that knowledge with the entire fleet. This allows individual skills to be reused, repurposed, and adapted across machines, factories, and tasks.
AI robotics startup seeks partners
Mbodi AI said it is rolling out production deployments to customers with help from partners such as ABB Robotics. This will bring its product into real-world facilities to automate tasks that were previously too complex or costly to program.
The company said its goal is “a world where robots are as easy to teach and adapt as software, turning them into true consumer commodities, and where automation and robotics become accessible to every factory and beyond.”
As part of its Y Combinator announcement, Mbodi is looking to connect with manufacturers, systems integrators, and robot makers to solve problems that were previously out of reach for robotics.
The company is initially focusing on pick-and-place applications. Chi said he sees potential for mobile manipulation and higher levels of autonomy.
“There will be specific smaller imitation models, which will need an intelligent system layer to coordinate them. The different models will be task-specific or based on a company’s own data set,” he said. “We see our system moving toward expertise-specific VLA [vision-language-action] models.