Training Robotic AI With Human Intuition

By Eli Samuel / April 9, 2019

The stealthy rise of collaborative robotics, in which robotic systems can safely interact with humans in the industrial workplace, may mean more moving parts in factories, but with a handmade touch. Instead of an automation technology imbuing a dystopic fear of job displacement, the dawning of trainable ‘cobot systems’ employ the strongest attributes of robotic programming and human intuition.

Founded in 2014, Micropsi Industries is at the forefront of teaching hand-eye coordination to robots using machine intelligence, requiring little programming expertise to operate large or small cobot systems. Through their Behavior Development Kit (BDK), a human-guided AI software named MIRAI, meaning Japanese for ‘the future’, trains robots with sensors, cameras, and deep neural networks that extract motion-relevant information to demonstrate even the smallest human movements.

Industrialization Through Demonstration

Amongst an AI boom of new automation technologies, affordable cobot systems could resolve an array of manual production problems — rarely overlooked, but impossible to automate until now —  in industrial manufacturing, from automating various complex tasks on factory floors to ensuring cables are plugged into the correct port.

Growing demand for faster, cheaper, flexible, and safer production is putting pressure on global manufacturers to reevaluate entire robotic systems that may be building their core business. Yet the transition to upgrade must happen quickly and accurately, or else manufacturers risk failing to meet production schedules. But whether it’s a dangerous, tedious, or extraneous task, MIRAI easily tackles each with simple instruction.

A philosopher and computer scientist by training, Micropsi co-founder and CEO, Ronnie Vuine, and his team created MIRAI’s vision-control system. The team inherently understood the inevitable benefits of training robots to learn directly from a live demonstration, rather than predetermined programming methods. Vuine, who currently splits his time between Micropsi’s headquarters in Berlin and New Lab in Brooklyn is building out a U.S. team after their recent round of funding.

Unlike in traditional systems, with MIRAI, no separate, manually prepared steps of pose estimation or edge finding are performed. Information flows directly from sensors to robot joints – through many neural network layers trained during demonstration.

“This is not science fiction,” Vuine insists, “but real robots guided by human expertise and intuitive movements. Micropsi Industries is a global business supplying actionable AI for manufacturing equipment around the world.” Micropsi intends to transform the industrial robotics market by applying MIRAI’s machine learning plasticity to improve the capabilities of human workers rather than replace them: “We automate highly repetitive tasks in factories where these workers usually get a better job immediately, and are typically grateful.”

Cobot See, Cobot Do

Ironically, most industrial robotics are much simpler (and clumsier) than popular science or science fiction suggests. Performing dynamic, non-precomputed movements is extremely difficult to program, and industrial robotics typically perform movements by only following and transitioning between fixed paths. However, Micropsi Industries’ MIRAI enables industrial robotic systems to learn organically from human instruction, where perception and precision is fundamental and natural.

Unlike traditional industrial robotics that can take weeks or months to become fully operational, MIRAI is virtually a plug-and-play AI software, controlled through a tablet controller. The entire process generally takes 48 hours, while the demonstration is often completed within a few hours. Instead of following pre-programmed trajectories, MIRAI-controlled robots react dynamically, duplicating movements it’s been shown moments before execution.

Learning a MIRAI skill means training a neural network to generate the right movements and forces when presented with a configuration of images and sensor readings.

After an initial demonstration of the preferred movement, guiding the robot by hand, externally mounted cameras and internal pose sensors collect data for learning on the MIRAI controller. Through the controller, image information from the cameras and sensors flow directly to robot joints, learned from the live demonstration. Once confirmed, MIRAI performs any demonstrated task autonomously.

MIRAI currently supports all Universal Robots models, ABB robots, and most robotic systems based on circumstantial hardware requirements. “AI applications also allow quicker customization for companies producing an array of products,” states Vuine. “The MIRAI software can apply to any robotic application without understanding much about the technology; it’s all physical, intuitive activity, not a month-long experience of learning or having to hire an outside expert or technical engineer.”

No-Brainer Results

The industrial robotics market is currently expected to grow to $71 billion by 2023,  which means more cohabitating cobots. Micropsi Industries is poised to leverage the industrial manufacturing market by automating various complex and dexterous tasks on the factory floor, but the extent of its application is yet to be realized globally. Micropsi’s target customers are primarily assembly factories, albeit the manufacturing of electronics, appliances, building materials, or tools. But while training robots prove to be a better practice than programming them, large-scale manufacturers have been slow to adopt comprehensive cobot systems. “Micropsi’s technology creates a solution for the challenges New Lab’s enterprise partners are facing,” explains New Lab CEO Shaun Stewart. “The exposure to collaborative robotic systems, and the opportunities to apply the technology to real problem sets in scalable ways is a game changer for large organizations.”

Financially, Micropsi’s MIRAI is an obvious, viable option for companies competing with larger manufacturers. One could even say it’s a literal no-brainer, as Vuine shares: “We want companies to design and deliver more consistent quality in their production systems. With MIRAI, customers can put robots and employees to work in entirely new fields, which will save time, cost, and unnecessary frustration in the future.”  

Ultimately, Micropsi Industries’ offers a disruptive AI technology that fills a critical gap in the robust robotics industry, empowering a whole new generation of reliable cobots to guide themselves while we lend our hands, eyes, and intuition elsewhere.

Are you an executive interested in understanding how robotics and frontier technologies are impacting the workforce? Contact New Lab’s team to learn more.

Close [x]
Sign Up for Our Mailing List
Receive the latest updates from our community and be the first to hear about New lab events and programming.