New ideas keep on emerging in robotics. But there are only a few that catch the attention of experts. Businesses are looking for more practical ways to train robots. That is why some startups are focusing more on new ways to enhance performance rather than just building large AI models. This is exactly what helped Mbodi get recognition.

Mbodi has received the top prize at the Automate 2026 Startup Challenge. The company is doing something different from its competitors. The co-founder, Xavier Chi, thinks that everyone is busy making large AI models. But there is a problem with this approach.

He said that the physical world is unpredictable. It has a lot of possibilities. When a large AI model goes wrong, it becomes hard to know what is happening inside it. The only option engineers can use is to train the model with more data and hope for the best.

Mbodi’s approach is totally different. Mbodi doesn’t believe in making giant models to handle everything. Instead, they build small and dedicated AI models. They pair them with agentic AI that is capable of taking actions on its own. Isn’t it interesting? Of course, it is. See how this works.

A person describes a specific task in plain language. The AI agent breaks this task into small steps. It then collects the information to train a robot. Now, that information is used to train robots. You know what it results in? This makes things easier and training faster.

Mbodi mainly works on cobots. These are also called collaborative robots that work alongside humans in industries.

Beating the Competition

So what made Mbodi win the competition? Well, their approach. They, instead of making one large AI model, build small agentic AI-focused models. It quickly got the attention of the judges at the Automate 2026 Startup Challenge. The company beat the nine other finalists to win the top prize. The judges were representatives from the world’s top tech companies, including Nvidia and Microsoft.

Chi said:

“Automate is a great show. It’s our favorite show, actually, one of the biggest in the whole robotic space, and we just thought we should go on stage and pitch our product or technology and see how people react to it. We got some really good feedback from people from all across different domains, so that’s a really great experience.”

Xavier Chi and co-founder Sebastian Peralta are ex-Google employees. They were not directly working in robotics, but they noticed a significant gap. They thought that despite all the progress in AI, there was still no reliable way to train robots. To fill that gap, they decided to start Mbodi and officially launched it in 2024.

Chi said they understand that, to scale up, they will need to set up several things right at the same time. This includes flexibility, capability, speed, throughput, and cost.

He said that one thing they made clear from day one was that they did not want to become a research company. He said that they did not want to stay in research mode for many years.

The main goal was to get technology that can actually benefit people and that companies can start using. Over the next year, Mbodi plans to focus heavily on deploying its system as widely as possible.

You know what’s interesting about the company? They started with picking and packing tasks. Now, the company says it has achieved an accuracy rate of 99.96%. And that is very significant for industrial robotics.

Another important thing is their technology. They have also changed it. The company says that earlier, they used a cloud-to-edge system. But due to excessive demand from customers, the company moved its vision system to edge devices. What does that mean? That means the entire system can now run locally. And there is no need for a cloud connection.

What’s more? Well, the company said that many pharmaceutical companies are also showing interest. They are seeing a lot of demand from them. They are running a pilot program with two of them. Though further details and the names of the companies have not been made public.