Standard Bots, an AI-native industrial robot manufacturer, recently raised $200 million at a $1 billion valuation, with customers including NASA, Amazon, and Lockheed Martin. The company builds robot arms for tasks like machine tending, welding, and assembly. Co-founder Evan Beard and Head of AI Leif Jentoft describe a stack built around a shared base model in the low billions of parameters, which customers adapt through demonstrations and fine-tuning.
The learned model handles perception, such as a zero-shot system for machine tending that locates and identifies parts, trained on over a billion images. Conventional programming handles motion and cell logic. Jentoft argues that data quality matters more than raw volume, and that in-situ interventions can fix edge cases with just a few dozen examples. Training happens in the cloud, but inference runs locally on edge GPUs, generating action chunks that stream to low-level control. Factories often lack reliable internet, so keeping the loop local is essential for uptime.
Standard Bots controls the full stack from arm to AI, which Jentoft says allows co-optimizing models and control policies. He claims no model today is truly hardware-agnostic, though Skild, a company pursuing general-purpose robotic intelligence, aims for cross-hardware generalization. Failures and human corrections from deployments feed fleet learning for customers that are not air-gapped, while defense customers in air-gapped environments send no data back. Simulation is used where possible, but tasks involving liquids, suction, or flexible materials remain difficult to reproduce in current simulators. The source notes that this approach prioritizes short-horizon tasks and production requirements over frontier-scale models. There are no other sources to compare or contrast with{