Lentil Robotics provides adaptable robot vision software that automates challenging, high-mix manufacturing tasks such as screw driving, assembly verification, and cable handling. The platform uses data-efficient, edge-native AI models that train locally in minutes and run on any standard industrial hardware, delivering robust results in fractions of a second without requiring special lighting or fixtures. This approach eliminates weeks of traditional vision pipeline tuning while maintaining accuracy for transparent, deformable, or reflective parts.
Funding
Funding not disclosed
Founders
Product
Problem
Traditional machine vision systems struggle with complex objects like transparent, deformable, or reflective parts, and require extensive manual tuning for every new SKU or variation. This makes automation impractical for high-mix production, where re-configuring classical pipelines takes weeks and inference speeds slow down cycle times, forcing manufacturers to rely on manual labor.
Solution
Lentil Robotics provides a hardware-agnostic vision software platform that automates challenging tasks with minimal engineering effort and low cycle times. The system uses data-efficient AI models trained locally on edge compute devices, requiring only a fraction of the data of conventional approaches and no internet connection. It performs object detection, pose estimation, visual servoing, and classification in fractions of a second, enabling robots to run at full speed. The platform is immediately robust to lighting changes, rust, dirt, and part variation, and can be deployed from hardware setup to executing a task in about 30 minutes.
Target Audience
Primary customers are manufacturing and industrial automation teams in high-mix production environments, including automotive, electronics, and general assembly operations that need to automate tasks involving transparent, deformable, or varying parts.
Features
- Edge-native AI models that run locally on standard compute devices with no cloud dependency or data leaving the factory
- Patent-pending, data-efficient training method that achieves production performance from a fraction of typical data collection
- Capabilities include object detection, pose estimation, visual servoing, classification, and assembly verification
- Hardware-agnostic design works with any standard industrial robot, RGB camera, and edge compute device without special lighting or fixtures
- Training pipeline completes in minutes, with full deployment from setup to execution in approximately 30 minutes
- Supports unlimited tasks and re-training under the software license, with multi-camera support at low incremental cost