The first of our three-part series of posts explored the obstacles preventing manufacturing from assimilating smart AI robots. The second post provided an ‘under the hood’ look on how an employment agency for robots can solve these challenges. We introduced the Industrial Robot as a Service concept of leasing a robotic workforce according to your production model, and highlighted the game-changing effects this model canhave on profits across industry.
In this post, we will describe what MusashiAI actually does: the robotic solutions we created and what led us to develop them.
It all starts with a meeting held a few years ago between Ran Poliakine, an Israeli serial entrepreneur, and Mr. Hiroshi Otsuka,President and CEO of Musashi Seimitsu Corporation, a Honda Motor subsidiary and Tier 1 auto parts manufacturer with 33 sites globally.
At the meeting, Mr. Otsuka raised two production challenges: first, an incredible 20% of workers carried out manual quality control inspection - a laborious and extremely costly process.
Workers require extensive training and qualifications, and applicants are few. In Mr. Otsuka’s eyes, inspection robots could replace humans to the benefit of both factories and workers.
A second issue was that an additional 20% of Musashi’s employees handle materials, moving them from one part of the production floor to another. Again, a case of people doing a tedious job that robots could easily replace.
Poliakine went home and did some research. He found out that visual quality control inspection was a pain point shared by many industries, but was especially acute when producing highly precise components with high nonconformity costs (for industries such as aerospace,automotive, and medical devices).
Material handling was surging across the board along with the booming demand for goods and consequentially Autonomous Mobile Robots have become a booming market.
Enter MusashiAI, the brain child of MR. Poliakine and Mr.Otsuka, offering a solution for each challenge.
The market has seen a proliferation of mobile robots for various applications, starting with AGVs (auto-guided vehicle) that move in preplanned routes and AMRs (autonomous mobile robots) equipped with built-in sensors so they can adjust their route and speed, making them more suitable for dynamic environments with human workers.
Their main drawback is set-up time and cost: the amount of sensors and computational devices on each vehicle makes AGVs and AMR sprohibitively expensive to buy, to maintain and to scale.
MusashiAI decided to take a different approach by introducing a proprietary and - literally out of the box – navigation solution.
We moved the robot’s ‘eyes’ and ‘brain’ out of the vehicle box to a central navigation and management system: a set of cameras installed on the facility’s ceiling, and the main control unit that assigns tasks to the robots, plans and optimizes routes, orchestrates fleet activity, and analyzes changes on the floor in real-time to adjust vehicle route and speed accordingly in the safest manner.
Not only does this lower the cost of autonomous vehicles dramatically, it actually enables converting a traditional material handling vehicle such a mechanical or electric forklift, cart, or a Pallet Jack, and even a cleaning machine into an advanced fully autonomous mobile robot. No need for expensive navigation equipment such as sensors, Lidars, cameras, and GPUs. Just plug it to the MusashiAI brain and go.
The MusashiAI solution also brings Autonomous Mobile Robots into production floors, warehouses, logistic centers, hospitals, airports,shopping malls, schools; and wherever autonomous vehicles are required to conduct tasks such as cleaning, disinfecting and material handling.
Scaling up or down is easy, as investment in each vehicle is minimal, and the robotic vehicle system is quick to set up and integrate.
The MusashiAI visual quality control inspection robot is a groundbreaking innovation, employing sophisticated optics, edge computing, and innovative machine learning to detect even the smallest surface defects in metal parts.
Robots doing quality control inspection is not a new concept. Robots use machine vision algorithms to detect defects, and AI-powered robots can detect small surface defects or faults in a fully assembled part.
However, the problem with using traditional rule-based machine vision algorithms to detect surface defects (as opposed to detecting defects in assembly parts) is the high false-positive rate they can generate, which makes robots an unviable option for a factory that produces tens of millions of parts annually.
Surface defects are achallenging beast to detect. In many instances, there is no clear criteria fordefects and it is very difficult to distinguish between the background surfaceand the defect itself - especially with unclear criteria.
This is exactly why human workers do the bulk of quality-control inspection for surface defects across many industries: the unparalleled combination of the human vision and the computational power of the human brain makes human quality-control inspection very difficult to replicate by a robot.
To train an AI-powered model, you normally need to feed thousands - sometimes tens of thousands - of images into the AI model. This is simply impossible for most manufacturing facilities, which operate in super high-yield. It would take months and years to collect the large sample size of bad parts required to train a traditional AImodel.
MusashiAI solves this challenge using, among other things, a different approach: instead of building on a set of known defects, we created a ‘Golden Sample Group’ to use as a reference for a flawless product.
The result is visual quality control inspection robots with the most precise defect detection rate to date.
Communication Center, Neve Ilan, Israel | 39-5 Daizen, Ueta-cho, Toyohashi-shi, Aichi, Japan