Neuromeka AI Lab · Preprint 2026
Learning-augmented robotic automation for real-world manufacturing
Neuromeka Co., Ltd., Seoul, Republic of Korea · * Corresponding authors: yunho.kim@neuromeka.com, joonho.lee@neuromeka.com
Abstract
Industrial robots are widely used in manufacturing, yet most manipulation still depends on fixed waypoint scripts that are brittle to environmental changes. Learning-based control offers a more adaptive alternative, but it remains unclear whether such methods, still mostly confined to laboratory demonstrations, can sustain hours of reliable operation, deliver consistent quality, and behave safely around people on a live production line. Here we present Learning-Augmented Robotic Automation, a hybrid system that integrates learned task controllers and a neural 3D safety monitor into conventional industrial workflows. It retains the core industrial automation backbone while introducing learning only where adaptability is needed. We deployed the system on an electric-motor production line to automate deformable cable insertion and soldering under real manufacturing constraints, a step previously performed manually by human workers. With less than 20 minutes of real-world data per task, the system operated continuously for 5 hours 10 minutes, producing 108 motors without physical fencing and achieving a 99.4% pass rate on product-level quality-control tests. It achieved near-human takt time with reduced cycle-time variability while consistently producing high-quality solder joints comparable to those of experienced operators. These results establish a practical pathway for extending industrial automation with learning-based methods.
Summary video
Watch on YouTube ↗Learning-augmented robotic automation
A factory-validated hybrid system that integrates reliable conventional automation with learning-based control in a safety-aware architecture. Instead of relying on a single end-to-end policy, our approach retains the core industrial backbone, an explicit task scheduler and pretaught motions for structured parts of the workflow, while introducing learning only where adaptability is required.
More details System framework and learning-based components
System framework
Overview of the learning-based components
Production-line deployment and validation
Watch on YouTube ↗
We deployed the system through continuous operation on a live electric-motor production line at the Neuromeka Pohang factory. Throughout the deployment, the system produced 108 motors over approximately 5 hours 10 minutes.
Across the run, the robot executed 324 cable insertion-and-solder operations. Two operations failed due to premature insertion termination, yielding a success rate of 99.4% per operation. Processed motors passed the downstream QC tests used in routine production: tensile and electrical tests. The system achieved near-human takt time with reduced cycle-time variability while consistently producing high-quality solder joints comparable to those of experienced operators.
Neural task controllers: adaptation and comparison with baselines
Watch on YouTube ↗
Neural controllers handled inevitable variability during the production process. They adjusted motion to handle randomness in motor pose, hole position, and cable bending.
The learned task controllers are compared with two learning-based variants and two conventional pipelines. Naive IL keeps the same imitation-learning framework but drops the structured image-processing pipeline and learns directly from full-resolution stereo RGB. VLA is a fine-tuned vision-language-action model (π0.5) with far more capacity and large-scale pretraining. The conventional baselines estimate poses with image segmentation and depth or RGB sensing and generate waypoints by rule; each applies only to some sub-tasks. Our controllers succeed in over 99% of trials on all three sub-tasks. The generic learning-based variants grasp the motor core almost as often, but seldom in the correct pose, and they need several recovery attempts per cable insertion where ours needs almost none. On this data budget, structured observations and task-specific inductive biases, rather than model capacity, are what make a learned controller reliable.
Efficient risk reduction with a neural 3D safety monitor
Watch on YouTube ↗The neural 3D safety monitor enabled safe human-robot coordination without physical fencing and reduced avoidable downtime. When workers approached to load materials or retrieve completed products, the system reduced speed or paused as needed and resumed autonomously.