Amazon Kiro

Amazon Kiro is a vision-language-action foundation model designed to give industrial robots the spatial reasoning required for complex manipulation tasks.

Kiro bridges the gap between high-level instructions and low-level motor control by processing multimodal inputs (RGB images and text) to generate precise 6-DoF end-effector trajectories. Built on a transformer architecture, it enables robots to handle unseen objects and unstructured environments without task-specific retraining. In benchmark testing, Kiro demonstrates a 25% improvement in success rates for pick-and-place operations involving occluded items. It integrates directly with AWS RoboMaker, allowing operators to deploy sophisticated autonomous behaviors across diverse hardware fleets using a single unified model.

https://aws.amazon.com/blogs/machine-learning/introducing-amazon-kiro-a-foundation-model-for-industrial-robotics/