A new upgrade to BrainOS’ autonomous navigation capabilities that enables faster decision making has resulted in major fleet operational improvements, including a 33% drop in human “assists” per route, according to Brain Corp performance data.

The upgrade, launched earlier this year, addressed feedback from the field that BrainOS-powered robotic scrubbers, despite outstanding overall performance, would sometimes appear to move hesitantly getting around obstacles, or navigating tight corridors within retail spaces. BrainOS is the company’s pioneering robotic AI software platform that now powers 16,000 robots worldwide.

“Some of the feedback was that in certain situations, the robots didn’t appear to move decisively enough or fast enough, even though they were doing their job just fine,” said Andrey Martsenyuk, Director of Technical Program Management at Brain Corp. “Appearances are everything, so it was something the team set out to improve.”

The project became known as MS2, short for Motion Stack 2, which made significant improvements to the robot’s AI and computational algorithms to enable faster decision making around obstacles, as well as in tight spaces. The development team leveraged Brain’s extensive real-world experience in public spaces – which now tallies more than 5 million autonomous hours of operations and 75 billion total square feet covered – to make the improvements. This, combined with improvements to the “perception stack” – i.e., how the navigation system uses the available sensor data to understand the world around it – enabled the robot to move faster and more confidently, while maintaining precision and improving recovery behaviors in challenging environments.

The results of MS2 have been impressive since being rolled out fleetwide earlier this year. The numbers:

• A 41% increase in miles per hour, which means the robotic scrubber is completing its routes that much faster on average without needing to increase the top speed. This enables robot operators to complete more cleaning routes within the same timeframe to achieve better cleaning efficiency.
• A 33% drop in the number of human “assists” per route. This enables robot operators to focus more deeply on other tasks without having to assist the robot to navigate through tight spaces.

The feedback from end customers has been extremely positive. They are pleased with the quantitative efficiency gains, as well as the way the robot now behaves in public: smooth, confident, faster and decisive. No hesitation at all – just consistent and reliable cleaning performance.

“It feels perhaps strange to think about whether a robot is moving confidently or not, but that’s the science of robot-human interaction, which Brain Corp takes very seriously. It impacts not only operational efficiency, but also how people perceive the robot,” Martsenyuk said. “If consumers feel unsure if the robot knows what to do, it impacts their experience as well.”