Eigen RadarAI
Archive
Analysis

Standard Bots keeps factory robot inference on site

Standard Bots executives have described how their industrial robots divide work between learned perception and conventional motion control. Training takes place in the cloud, while inference runs on local hardware to support operations where internet access is unreliable. Real demonstrations and human corrections inform adaptation; customers disconnected from outside networks do not return learning data.

Artificial Intelligence··Midday
An industrial robot arm holds a metal workpiece before an open machining-cell fixture, with a wrist sensor and closed hardware box nearby.

The model finds parts; software controls motion

Standard Bots, an industrial robot-arm maker, has described its AI architecture through executives Evan Beard and Leif Jentoft. In the October 10 technical interview, they outline a division between learned perception and conventional programming for factory tasks such as machine tending, welding and assembly.[1], [2]

For high-mix machine tending, Jentoft says the model identifies and locates the parts that an operator asks it to find. Conventional programming then handles motion and cell logic. Beard says customers adapt a shared base model through demonstrations and fine-tuning, rather than receiving an entirely separate model for each factory.[1]

Training is in the cloud; inference stays local

Training takes place in the cloud, while inference runs on local hardware. Wrist cameras and other sensors send data over internal gigabit Ethernet. Edge graphics processors generate action chunks for low-level control. The executives attribute the local operating loop to unreliable internet connectivity in factories and warehouses, where continuous operation matters.[1]

Human corrections feed customer-specific learning

Liquids, suction and flexible-material cutting are difficult to reproduce in current simulators, so the company uses real demonstrations and human corrections. Data from customers disconnected from outside networks does not return; other fleet-learning arrangements differ by customer. Its StandardOS developer platform offers application programming interfaces and software development kits for connecting outside models, currently requiring integration code. Easier collection, training and deployment workflows remain company plans.[1]

References

  1. News sourceLatent.SpaceStandard Bots details local inference for factory robots↩1↩2↩3↩4
  2. News sourceEl Ecosistema StartupStandard Bots details local inference for factory robots↩
AI safety and regulationChip wars and the supply chain

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →