Among the robots responding to the novel coronavirus crisis are disinfection systems using ultraviolet light to kill pathogens. However, UV radiation can also be harmful to humans, so such robots need to operate autonomously, or they cannot work in spaces such as hospital rooms if people are present. Akara Robotics Ltd. has released Violet, a disinfection robot that can sense people and turn off its emitter if necessary.

The Dublin-based startup spun out of Trinity College and has developed Stevie, a social robot for nursing homes, said Conor McGinn, co-founder and CEO of Akara Robotics.

“We’ve been exploring a disinfectant application for the last year or so, and we came together and decided to develop a dedicated UV robot” he told The Robot Report. Violet is the latest robot addressing the COVID-19 pandemic.

Movidius optimizes vision processing

“With Violet, we focused on learning models, robotic navigation, and computer vision,” said Niamh Donnelly, co-founder and director of AI at Akara.

How is Violet’s vision different from that of other mobile robots? “There are two parts. The first is that computer vision is normally on a power-hungry CPU or GPU,” explained Jonathan Byrne, data scientist, senior software developer, and chief drone pilot at Intel Corp.‘s Movidius unit. “Movidius is low power, which lends itself to robotics, which needs a powerful processor. The whole stack might be easier to program, but it’s difficult to keep at low power.”

“Movidius includes optimized cores and a pipeline that can pull a frame into RAM processing with 10 reads from memory on a dedicated CPU,” he said. “Movidius processes line by line, in parallel, and hits with many filters at once.”

“Since Myriad X, we now have an embedded chip system dedicated to deep learning, before the perceptual stack. We can optimize multiple networks to run simultaneously on the hardware,” Byrne said. “This is useful for tracking, identifying people, or detecting whether they’re happy or sad.”

Designing Violet for collaborative cleaning

“There are a number of factors to consider for a UV disinfection robot, from a design perspective,” said McGinn. “The first is the total power of light, where intensity follows an inverse square law. You either need powerful lighting rigs to cover a room or intelligent navigation with a lower-powered light.

“Violet can move more and get closer to surfaces — most go to two or three places within a room; we go to four or five,” he said. “The effectiveness was verified by an independent microbiology lab.”

“The second point is safety — it’s important to categorize whether someone is patient or visitor, who is not likely to be the first person to share a room with a robot,” McGinn said. “More likely, it would be the cleaning or medical team, which is more familiar with standard operating procedures and wearing personal protective equipment with specifications for occupational exposure limits.”

A robot operator could remotely control Violet, or a cleaner could even work alongside a robot, McGinn said. “We’ve been working in hospitals with staffers on procedural elements, such as what combination of UV and friction [human scrubbing] is needed,” he said. “Only then do we bring in artificial intelligence, which provided the opportunity to use computer vision and person tracking to turn lights on and off.”

“Not only can the robot direct where the light shines — every item has to withstand UV, which CT scans cannot — but we can also actually control what areas of the room are disinfected with UV-C,” said Donnelly. “Violet has a 360-degree field of view, so it can detect exactly where a person is in a room and shield them from radiation.

“By maneuvering more, we were able to go into the radiology department, which had previously considered robots but decided against them because they couldn’t get behind the CT scanner, which left a huge shadow. Our design decisions were defined by the users,” she said.

Working on fleets, and robots in radiology

Most disinfection robots so far are operating in small numbers, but could fleets of mobile robots work in hospitals or other spaces? “Before we manage multiple robots, we want to test as much as possible first, starting with one robot per room,” said McGinn. “As we expand beyond radiology, doctors have come down to ask questions.”

“We will have end-to-end management for a fleet,” said Donnelly. “We’re thinking now about scaling up.”

“There is currently a lot of interest in infection control in hospitals and nursing homes, and as the new normal demands robots elsewhere, having a fleet of five or 10 smaller and more versatile robots might be useful,” McGinn said. “A second issue is that it’s very hard to clean rooms in radiology, with a million-dollar unit in the middle of the room, you can’t use air-based sprays. There’s half an hour to an hour between sessions where the room is not being used. Ten minutes of time is worth $300, so we could reduce that time with more robots.”

“In public transport, it would make sense to deploy more robots,” he said. “Before we start committing to a design, we want to know what different users want.”

Service robot experience

Several service and social robot providers have pivoted to healthcare applications, but Akara Robotics is keeping its commercial focus. “We see this as a reprioritization,” said McGinn. “We still have the intention to get back to Stevie, which has real potential.”

“We had two challenges — the first was market fit,” he said. “It took time to bring our robot into nursing homes, which never had robots before. It took long-term pilots to train and integrate where the robot might be most useful. We already have partnerships on three continents.”

“The second challenge was operations,” McGinn said. “We want to require fewer people to deploy a robot, and there’s always more to do in adding functionality. With Violet, we focused mainly on quickly building the operational piece.”

“To design and build a social robot, we have conversations with groups of people,” added Donnelly. “With Violet, navigation was a largely solved problem but still quite difficult. We reused technology and open-source software, and we got help with integration from the Movidius team.”

“We’ve learned a lot about our own technology from collaborating with startups,” said Intel’s Byrne. “Finding a value add like making Akara’s robots safer is good application.”

Deploying Violet

“We’re still working hard to get robots in hospitals,” said McGinn. “It will take a while longer to mass-produce, but in the short to middle term, we’re getting Violet proved, and we’ll deploy on a day-to-day process in a couple of months.”

“A whole range of industries getting back to work needs infection control, such as airports, rail, retail, and construction,” he said. “We have a couple of projects with longer horizons.”