Smart Machines & Factories
Teaching old robots new tricks
Published:  13 December, 2017

Two start-ups in the US are threatening to the shake up the robots market by turning conventional industrial robots into collaborative machines, and by my making it easy for anyone to program a robot, without needing to be an expert. Tony Sacks reports.

Industrial robots are continuing to make remarkable inroads into factories in most parts of the world (the UK being a notable laggard). The International Federation of Robotics (IFR) predicts that the number of robots installed around the world will climb by at least 15% a year for the next few years so that, by 2020, more than 1.7 million new industrial robots will be in operation, taking the total in service above three million.

But there a several factors that are holding back robots from making even more progress. One is the need for experts to program the robots to ensure that they perform their complex three-dimensional movements safely and effectively, without adversely affecting other factory operations.

While the issue of safety is being tackled by the emerging generation of collaborative bots (cobots) which are designed to operate alongside humans without needing protective cages, these robots tend to limited in their lifting capacities and operating speeds. At present, cobots still represent a fairly small part of the industrial robotics market.

A couple of start-up companies have been formed in the US with the aim of tackling some of these issues. One, a Massachusetts-based business called Veo Robotics, which was founded in 2016, is creating “intelligent, human-aware systems” for industrial robots. Its mission is to bring “perception and intelligence to industrial automation so that humans and machines can work closely together”.

Veo’s first product uses a new class of intelligent algorithm running on parallel computing hardware – used in combination with distributed 3D sensors – to allow humans to work safely alongside large, powerful industrial robots. In effect, Veo’s technology will turn traditional, heavy-duty industrial robots into human-friendly cobots.

The Veo system works in conjunction with the robot’s standard controller, overriding its operation if it senses potential danger to nearby humans. Four depth-sensing cameras are placed around the robot’s work area. Items within this area are designated as workpieces or forbidden areas. If something – such as a human – moves into the area, or if there is some other deviation from the norm, the system can slow down or stop the robot. The system will bring the robot to a halt if it is not 100% sure that it is safe to operate the arm – for example, if the view from one of the four cameras is obscured, or if the system cannot see behind a large workpiece.

“This is a revolutionary time for global manufacturing," says Patrick Sobalvarro, Veo's CEO and co-founder. “For more than 100 years we've been using machine tools that are unaware of their users and unable to respond to them intelligently.”

He believes that there is now an opportunity to make a step change in manufacturing to create fluid, intelligent, highly flexible and productive environments where humans and machines work closely together, each doing what they do best.

Veo's technology will allow industrial workers to work collaboratively with machines, bringing more flexibility, productivity and precision to manufacturing.

Veo has recently attracted $12m in funding from investors including GV (formerly Google Ventures) and Next47, a venture firm set up by Siemens.

The funding will allow the company to accelerate its hiring of personnel in areas such as computer vision, engineering, product management and systems engineering, as well as supporting product development.

The team behind Veo Robotics has a combined 50 years of experience in robotics, artificial intelligence, sensor perception and industrial automation. CEO Patrick Sobalvarro is a serial entrepreneur who was formerly the president of the cobot pioneer, Rethink Robotics.

“For decades we've been told robots were to blame for the dearth of manufacturing jobs in the US, but that's about to change,” predicts Bilal Zuberi, a partner at Lux Capital which led the latest round of fundraising. He believes that the team behind VEO has “a counter-intuitive vision for the future: if you improve the robots, manufacturers will be able to hire more people for better jobs.

“Veo is giving industrial robots the gift of perception and intelligence so these helpful machines can still do the dangerous heavy lifting, yet safely work alongside humans.” Bilal adds.

Veo is currently working with customers to apply its technologies in sectors including automotive and consumer packaged goods, as well as with appliance manufacturing and automated distribution. It is planning several pilot trials with these customers over the coming months.

Even if Veo does bring collaborative abilities to heavy-duty robots, programming them will remain a challenge, usually requiring experts with considerable experience. This is where the second American start-up comes in.

Embodied Intelligence, located on America’s West coast, is harnessing VR (virtual reality) and AI (artificial intelligence) to allow almost anyone to program industrial robots.

The team behind Embodied Intelligence has a combined 30 years of experience in AI, deep learning and robotics. President and chief scientist Pieter Abbeel has pioneered many recent breakthroughs in robot learning at his laboratory at the University of California Berkeley. These range from robots that can learn vision-based manipulation from their own trial and error, to automated systems for organising laundry.

Embodied Intelligence is developing AI software that can be loaded onto existing robots. While traditional programming of robots requires writing code – a time-consuming process, even for robotics experts – the new software will allow anyone to program a robot simply by wearing a VR headset and guiding a robot through a task. This demonstration will train deep neural nets, which are then further tuned through the use of reinforcement learning, resulting in robots that can be easily taught a wide range of skills.

The technique can be used to train robots to perform complicated tasks such as: picking parts out of cluttered, disorganised bins; completing assemblies where conventional automation struggles due to variability in parts and configurations; and manipulating deformable objects such as fabrics, wires and food.

“Traditional robot programming requires substantial time and expertise,” Abbeel explains. “What we will provide is an AI layer that can be added to any existing robot, enabling robots to learn new skills rather than requiring explicit programming.”

Embodied Intelligence’s CEO, Peter Chen, adds that “VR devices provide an easy way to control and teach physical robots. Since the robot simply mimics the hand motion that’s tracked by VR, a person without any special training can make the robot do the right thing right from the beginning. The robot will keep learning and, after a while, it says, ‘I got this, I can do this task on my own now.’”

Like Veo, Embodied Intelligence has also recently completed a successful funding round, raising $7m in seed funding (from a group that also includes Veo backer, Lux Capital). The money will be used to help develop the AI software.

The fundraising was led by at Amplify Partners. Sunil Dhaliwal, a general partner in the firm. points out that “recent breakthroughs in AI have enabled robots to learn locomotion, develop manipulation skills from trial and error, and to learn from VR demonstrations. However, all of these advances have been in simulation or laboratory environments. The Embodied Intelligence team that led much of this work will now bring these cutting-edge AI and robotics advances into the real world.”

“Despite remarkable advances, robots continue to require armies of PhDs to make them perform human-like tasks,” adds Shahin Farshchi, a partner at Lux. “Embodied Intelligence advances the state-of-the-art in reinforcement and imitation learning, yielding robots that can be trained to do complicated tasks, quickly.”

“Advanced robotic capabilities have been confined to established players that can afford costly R&D efforts,” says Embodied Intelligence’s chief technology officer, Rocky Duan. “Our teachable robots will empower any size businesses to incorporate robotics into their manufacturing processes and keep up with the competition.”

For further information please visit: