Close the automation gap in manufacturing with robots that learn on the job. Learn how Walden turns every correction on the factory floor into training data.
Walden Robotics is no stranger to the factory floor. Its founders spent about a decade at Toyota working inside the Toyota Production System before spinning Walden out of the Toyota Research Institute in January 2026. Today, Walden’s general-purpose robots work alongside people on a Toyota production line in North America, and Toyota, Boeing, and Samsung are strategic partners.
Automation Gaps in Manufacturing
Manufacturing has an automation gap. Machine tending, parts kitting, tool setting, and sub-assembly work change too often for fixed automation, so skilled workers still handle them by hand. There are half a million open machinist positions in the US. Walden’s robots work alongside people as assistants, taking on the repetitive parts of the job so workers can focus on the craft and on improving the process.
Walden’s bet is that general-purpose robots can close that gap, and that the most valuable data in physical AI comes from robots learning in a real production setting, not in the lab. That means earning a place on a live production line before the robot is perfect, under the same rules every plant runs on: safety, quality, delivery, and cost, in that order.
To make that bet pay off, Walden needed robots that could:
- Hit customer targets for success rate, autonomy, and cycle time from day one, on lines where a car can roll off every 54 seconds
- Keep up with parts, line layouts, and standard operating procedures that change from week to week
- Maintain performance when lab-trained models meet a real factory floor
Solution
Walden’s relationship with CoreWeave goes back further than the company itself. Adrien Gaidon, Walden’s co-founder and CSO, was the first-ever customer of Weights & Biases, now part of CoreWeave. When Walden launched with $300 million in funding, CoreWeave Ventures was among its investors.
Today, Walden brings the robots and the factory know-how. CoreWeave brings the tools and physical AI expertise to help it scale.
Humble robots that ask for help
When a robot is about to do something it shouldn’t, it stops and calls a remote assistant instead of pressing on. It’s applied jidoka, the Toyota Production System principle of a machine that stops itself and calls a person, applied to physical AI. Gaidon calls them “humble robots.”
Every correction becomes training data
A remote assistant’s correction shows the robot the right move at the exact moment it was about to go wrong. Walden treats these on-policy corrections as its most valuable data and feeds them into the next training run. The loop from field data to an updated policy takes weeks, and increasingly days.
Customers write the tests
Customers bring the requirements: success rate, percent autonomy, mean time between interventions, and cycle time. Walden builds those metrics into its simulation and pre-deployment evaluations. Before it ships an over-the-air update, it confirms that those results match what happens on the line.
Experiments tracked with CoreWeave Forge
Robot learning is “very empirical in nature,” Gaidon says. Walden runs constant experiments with Weights & Biases Models on data mixtures, models, and simulations, and continues to work closely with the CoreWeave team to track and compare them.
An investor and partner from the start
Along with its investment, CoreWeave brings physical AI specialists, including mechanical engineers and robotics experts, to work with Walden’s team.
“They’re an investor in us. They’ve been an early believer. We’re starting our journey on physical AI together.”
Adrien Gaidon, Co-founder and CSO, Walden Robotics
Proof from the Production Floor
Walden measures success the way its customers do. Customers like Toyota, Samsung, and Boeing arrive with long lists of metrics, and Walden has to keep passing them to keep its place on the line.
On a production floor in two months
Walden was founded in January 2026. Within two months, the team built its first two robots, installed them on a production floor during a non-production weekend, and passed the customer’s gates. Since May, the robots have worked full shifts alongside people.
100% task success on 8-hour shifts
For months, at one of the largest manufacturing plants in North America, Walden robots completed every task they were given, with remote assistants stepping in when needed. One job is machine tending: picking camshafts from supplier dunnage and placing them with millimeter precision into V-blocks on a moving conveyor.
Robots that keep up when the line changes
Production engineers continuously update procedures and move work between lines. Walden’s robots learn from those changes instead of waiting to be reprogrammed. When people on the floor ask whether the robot could take on another task, Gaidon says the goal is “to say yes as fast as we can.”
