Twitxy Saved

Chef Robotics

@ChefRobotics

Empowering humans to do what humans do best. We're hiring: chefrobotics.ai/careers
San Francisco, CA chefrobotics.ai Joined May 2018
2 Following 1.5K Followers 338 Posts

Latest posts 5

Chef Robotics @ChefRobotics
A robot doesn’t need to do an entire human job to be useful. It needs to do a specific task exceptionally well.

Our CTO and founding engineer, Luis Rayas, joined the FuturePulse podcast to explain how this shapes our work at Chef.

For food assembly, that means fast, precise, consistent portioning, freeing people to solve problems and handle unexpected situations on the production line.

This also shapes how we think about general-purpose robotics. An excavator’s shape makes sense for digging; giving it a human body wouldn’t make it a better tool. Luis sees more promise in a general-purpose robot brain that learns to manipulate different things, with hardware suited to each application.

At Chef, deploying robots teaches us which problems actually need learning and where a simpler approach works better. Our growing fleet also generates data that makes new learning approaches possible.

That’s what “deployment first” means at Chef: customer needs shape the technology, and production experience shapes what we build next.

Thanks, Melica Moshiri, for having Luis! Watch the full conversation: youtube.com/watch?v=OhEATM…

#PhysicalAI #foodautomation #AIandrobotics
www.youtube.com Beyond the Humanoid Will we ever build a general-purpose robot? Chef Robotics’ CTO argu...
Chef Robotics @ChefRobotics
Working with a large customer can give a robotics company the funding, production experience, and orders it needs to grow. But what happens when that customer slows down or finishes its rollout?

The Robot Report explores this question through Symbotic’s relationship with Walmart. One thing that stood out to us was the recommendation to involve several customers early in product development so their requirements help shape your system from the beginning.

In food manufacturing, for example, manufacturers may work with different ingredients, portion sizes, and production line layouts. Understanding those differences early helps a team decide what to build into the core system and what to make configurable.

Early deployments also show how much engineering and on-site work each new customer requires. Knowing that helps us plan how to serve new customers while supporting existing ones.

Read Deepak Jayaraj's article to learn more about managing customer dependence and planning for growth: therobotreport.com/what-symbotic-…
Chef Robotics @ChefRobotics
Sam Padilla and Varun Nair's robotics map gave us a new perspective by depicting the robotics industry as a subway system. Subway stations represent the industry's layers (e.g., robots, models, deployments, data services). Companies that span more than one layer are placed on one or more subway lines between the stations. The busiest connection, not unexpectedly, is the deployments ↔ robots line.

Some companies, like Dyna Robotics, ride more than one subway line, and NVIDIA is represented at four different subway stations.

Beyond the visual representation of how different layers of the robotics industry are connected and which companies are at the forefront of each, we're interested to see how these layers and connections will change over time and how the pendulum between robots and deployments, models and simulation, data tools and data services will swing over time.

Explore the map and let us know what you find interesting about it: roboticsmap.org/c/chef-robotics
Chef Robotics @ChefRobotics
Part 1 of our Food World Model blog series explored how our model predicts the weight of a candidate pick and the surface it will leave behind in an ingredient pan. Today, we’re taking a look at where our training data comes from and how we turn production robot data into a scalable training flywheel.

Getting the training flywheel right is crucial because with world models, it's not enough to predict a plausible future. A world model's predictions need to hold up against physical measurements and, in our case, across a range of food ingredients and production conditions.

Our AI research engineering team turned 150 million production robot actions into a reproducible data corpus more than 10x larger than the one mentioned in Part 1, and each new training run is cheaper than the previous one.

Read more: chefrobotics.ai/post/building-…
Chef Robotics @ChefRobotics
Taking the PACK EXPO conversations out on the water! 🛥️

We hosted a boat cruise in Chicago during PACK EXPO to bring food manufacturing folks together off the busy show floor and spend more time getting to know one another.

Trade shows bring so many people into the same city, but there’s rarely enough time for every conversation. We’re glad we could make some room for that.

Thank you to everyone who joined us and made the cruise part of their time in Chicago. It was a pleasure having you aboard!

If we didn’t get to connect in Chicago, we’d love to continue the conversation: chefrobotics.ai/contact-us?utm…

#PACKEXPO #foodautomation #physicalAI #foodmanufacturing