Walden Robotics' deployment-first playbook for general-purpose robots
Key points
Key takeaways from a theCUBE interview with Walden Robotics co-founder and CSO Adrien Gaidon, recorded live at CoreWeave Fully Connected conferencein San Francisco in Octoberer 2026:
Deployment first. Walden builds general-purpose robots for manufacturing and prioritises getting them onto real production floors fast, showing usefulness and ROI from day one, then compounding it. Once deployed, factory teams keep finding new tasks, and the answer to "could it do that?" is "yes, as fast as we can", not "we will get back to you in a year".
The automation gap. Walden targets machine tending, parts kitting, tool settingand subassembly, tasks that have resisted automation for years. With roughly half a million open machinist positions in the US, its robots work alongside people, handling the parts of those jobs that do not need human craft, letting people focus on craftsmanshipand process improvement.
Variance is the next frontier. Lean practice reduces variance, but the next step of mastery for manufacturers like Toyota, Boeing and Samsung is improving processes while machines continuously learn in deployment, raising mixand personalization without sacrificing safety, quality, delivery or cost, in that SQDC order.
From TRI to the factory floor. Founded in Januaryr 2026 as a spinoff of Toyota Research Institute,where CEO Russ Tedrake, Gaidonand colleagues built the Large Behaviour Model project, Walden put its first two robots on a production floor within two months. It has been doing real work alongside people since May, working towards full shifts.
The funding. Walden exited stealthin Julyr 2026 with a $300 million seed at a $1.1 billion valuation, co-led by Toyotaand Deviation Capital, backed by NVIDIA, Boeing, Samsung Venturesand others,with CoreWeave also an early investorand infrastructure partner.
The CoreWeave loop. CoreWeave brings physical-AI specialists, mechanical engineersand robotics experts, plus infrastructure for the full run, curate, improve, evaluate loop,with much of a robot's learning rehearsedin simulation before it reaches the floor.
Trust is the bottleneck. Capability is progressing fast, but what limits physical AI is deployment,building customer trust gradually, proving it with data on live production lines. Autonomy is not binary but a ratio, where continuous metricsand thresholds unlock new capabilities. Safety stays paramount,and evaluation happens in simulation first, using tools includingthe Drake simulator Russ Tedrake built at MIT.
The embodied stack. Gaidon,a 20-year computer vision veteran, traces the field from the deep learning revolution through LLMs, VLMs, VLAsand world models. Frontier modelsare good enough to demonstrate value, then co-designed systemswith humans in the loop keep improving in deployment.
Read more: theCUBE on YouTube · Walden Robotics
