RIOS raises funding, emerges from stealth with AI driven dexterous robots

The Robot Report
RIOS lab automation

DX-1 is able to recognize and handle parts in multiple orientations. Source: RIOS

RIOS Corp. today emerged from 18 months of “stealth mode” and announced that it has raised $5 million in venture funding. The startup said it has developed “highly dexterous robots that handle hard-to-automate tasks in unstructured environments.”

Palo Alto, Calif.-based RIOS spun out of Stanford University and was founded in 2018 by former Xerox PARC engineers. The company claimed that not only can it automate individual lines with its artificial intelligence and manipulation systems, but it can also build “lights-out” factories with its network of systems integrators.

“We’re building robots of the future — ones that can learn on the job, construct models of the world, and extend these models to perform different tasks,” stated Dr. Bernard Casse, founder and CEO of RIOS.

RIOS tackles complex tasks

“Our sweet spot is complex automation — i.e., automation in unstructured environments that is hard to automate and that typically requires human-level dexterity,” Casse told The Robot Report. “By ‘hard to automate in unstructured environments,’ we mean handling hundreds of SKUs, precise manipulation and assembly, sorting and selecting objects coming at different positions on an assembly line, etc.”

“If a task on an assembly line cannot be currently automated by any of the ‘big four’ robotics companies [ABB, FANUC, KUKA, and Yaskawa] and would require millions of dollars in customized engineering solutions that would require an overhaul of the factory, this is when RIOS comes in,” he said. “Our robot can be dropped into an existing assembly line to perform many types of tasks.”

“With a first-of-its-kind haptic intelligence platform combining dedicated hardware, computer vision, and AI, our robots can learn to grasp and handle many types of objects, precisely selecting and assembling parts and performing complex manipulation tasks,” he said.

Casse cited the example of assembling an oil filter (see video below). “We provide the robot with three to four components, and it takes them and reassembles them,” he said. “It doesn’t matter the orientation of the parts or their physical location; it will still manage to assemble them.”

RIOS builds robotic eyes, brains, and hands

RIOS said it has developed and deployed a “full stack” of perception systems, centralized AI, and a purpose-built end effector using advanced tactile sensors. However, it involved more low-level work than expected, said Casse.

“We had to do a lot of ‘plumbing’ work that we didn’t anticipate — stuff we thought existed and that we could buy off the shelf turned out to have never been created,” he said. “And elements we thought would work out of the box tended not to work out of the box! We had to write drivers for hardware, fix a lot of bugs in ROS [the Robot Operating System], hack robots to get real-time control, develop our own data-processing platform, etc.”

The company said its first product, DX-1, is a multipurpose robot that can interact with objects in both static and dynamic applications, such as bin picking and moving conveyor belts.

The DX-1 includes thousands of miniature sensors in a 2-in. by 1-in. footprint. How does that compare with other robot sensors? “If you’re talking about pressure sensitivity, then you can simply blow on it, and it will detect it,” Casse replied. “It also has a large dynamic range. Even if you hammer on it, it won’t saturate and reconstruct your pressure profile, which is unusual for sensors. The industry standard is well below that.”

“But we’ve developed way more capabilities than this,” he added. “First, our robot is biomimetic, which means that it can detect temperature, frequency, and texture. The performance for all these sensing modalities are what you would expect from a human hand. We’ve added proximity sensing as well. But most importantly, there are proprietary algorithms that synthesize and process the data to give the robot capabilities like optical grasp, slip detection, surface topography mapping, and texture discrimination.”

RIOS haptics

RIOS said its end effector is built for sub-2ms reaction time and sub-0.1mm tolerances.

DX-1 demonstrates flexibility in pilots

In pilots, DX-1 quickly learned to adapt to multiple use cases in manufacturing, automotive assembly, lab automation, supply chains, and food services, said RIOS.

Most industrial robots are scripted to perform single repetitive tasks on tightly controlled assembly lines. “Our robots possess higher levels of dexterity, cognitive skills, and autonomy than today’s robots,” said Casse. “Right now, it takes about one to two hours for DX-1 to learn a new task from the ground up, including learning new objects in the scene.”

What robot arms does DX-1 use? “For lab demos, we’re using Universal Robots,” said Casse. “For pilots and deployments, we’re currently partnering with FANUC. We’re also close to partnering and supporting Yaskawa robots for pilots and deployments.”

“For deployment, we need to provision the system, which takes longer, because there are workflow considerations, corner cases, possible failure modes, etc.,” he said. “But we’ve shown to customers and prospective customers that the robot was able to do the task at hand in this timeframe.”

“Everyone who’s seen our DX-1 robot in action is amazed at what it can do,” Casse said. “We had feedback from customers saying that the big four robotics companies said their tasks could not be automated, but we’ve demonstrated that DX-1 can do a lot of these tasks.”

RIOS raises funding, emerges from stealth with dexterous assembly line robots

DX-1 can recognize and grasp previously unseen objects in this e-commerce application. Source: RIOS

Investing in growth

Menlo Park, Calif.-based Valley Capital Partners (VCP) and Morpheus Ventures led the funding, with participation from Grit Ventures, Motus Ventures, MicroVentures, and Alumni Ventures Group. Japanese strategic investors Fuji Corp. and NGK Spark Plug Co. also contributed. VCP Managing Partner Steve O’Hara, Morpheus Principal Howard Ko, and Grit Managing Partner Jennifer Gill Roberts have joined the RIOS board of directors.

“While others in the space are tackling simpler use cases like pick and pack, RIOS has combined its haptic and perception capabilities to deliver a next-generation solution today,” said Ko. “The RIOS robot does complex autonomous manipulation that enables it to easily integrate into existing assembly lines to deliver a real-world increase in productivity and cost savings.”

“We have nine full-time staffers,” said Casse. “It’s engineering-heavy — 75% engineers — since we had to build the full stack from the ground up. We work with about a dozen contractors to augment our workforce. Business processes are mostly outsourced; we have three business consultants including a firm. We will soon hire more business professionals.”

RIOS offers RaaS

RIOS delivers DX-1 via a robotics-as-a-service (RaaS) model, charging a flat monthly fee that includes installation, initial programming, regular software updates, re-programming as needed, maintenance, and 24/7 monitoring. The company said the flexible model eliminates upfront capital costs and allows facilities to add automation quickly and cost-effectively.

“We believe in Bernard’s vision for automation as a service, recognize the potential for wide-scale global deployment of RIOS’s dexterous robots in many environments, and have seen firsthand how the company’s highly intelligent robots can manipulate many types of objects with the incredible dexterity of a human hand,” said Steve O’Hara, managing partner at VCP.

“We can retrofit our system to the customer’s site,” said Casse. “We work closely with customers to make sure that we do not interrupt the workflow in their factories. We also work with our network of systems integrators to make sure we can provide a continuous workflow, even beyond the line. We’re ready to take on more use cases and customers.”

The post RIOS raises funding, emerges from stealth with AI driven dexterous robots appeared first on The Robot Report.

Source: therobotreport

Leave a Reply