Standard Bots keeps factory AI local and focused on specific jobs

Standard Bots says its industrial robot arms use AI to identify parts, while conventional programming handles the motion and surrounding cell logic. The approach offers a practical look at how AI is being adapted for factory work, where reliability and uptime matter more than a model’s parameter count.

Watch Desk analysis

What happened

In an interview published by Latent Space on 10 October, Standard Bots co-founder and chief executive Evan Beard and head of AI Leif Jentoft described the company’s approach to industrial robotics. The company makes robot arms for jobs including machine tending, welding and assembly.

Jentoft said its machine-tending system can identify parts without task-specific training, using a backbone trained on more than a billion images. The model handles perception, while conventional programming controls motion and the rest of the robot cell. Beard said training happens in the cloud, but inference runs locally at customer sites.

Why it matters

That split is a useful counterpoint to the idea that a more capable robot must be driven by one all-purpose model. Standard Bots keeps the learned part focused on identifying what is where, then relies on programmed behaviour for movement and cell logic. The company says its approach is designed for short tasks with production requirements for cycle time and reliability.

Local inference also addresses a straightforward factory constraint: Beard said many sites lack reliable internet, while downtime can undermine customer acceptance. The company’s account is not an independent performance assessment, but it gives a concrete picture of how one robotics maker is fitting AI around real operating conditions.

Our read

The interesting bit is not the billion-image figure on its own. It is the division of labour: use learning where perception benefits from varied data, and conventional programming where a production process needs predictable control. That is less cinematic than a humanoid robot, and rather more like the work factories actually need done.

For developers, the practical lesson is to start with the task and the data it needs, rather than assuming a larger model is the answer. The company says it uses deployment failures and human corrections to address edge cases, although data-sharing arrangements vary by customer.

What to watch

• Whether Standard Bots’ division between learned perception and programmed motion proves reliable across more factory tasks.
• How its planned tools for collecting data, training models and deploying them to robots develop.
• How customers handle deployment data, particularly in air-gapped environments.

Discussion spark: For factory robots, should teams keep AI narrowly focused on perception and leave movement to conventional programming, or is that division likely to limit what the machines can handle?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.