For three years, a group of robots, known as CoBots, has been navigating the corridors of Carnegie Mellon University’s Gates and Hillman centers and Newell-Simon Hall, running errands and guiding visitors without human supervision. On Nov. 18, their collective odometer reached 1,000 kilometers — more than 620 miles — a first-ever accomplishment for indoor autonomous robots.
And they have no plans to stop.
Manuela Veloso, professor of computer science and founder and leader of the CoBot project, said reaching 1,000 kilometers is a scientific and engineering milestone that clearly separates common technology demonstrations from a long-term operation, offering new solutions and challenges for artificial intelligence and robotics researchers.
“We were enthusiastic and daring to set a research agenda to include the actual long-term deployment of autonomous mobile robots in our real building environments,” Veloso said. She explained that the research addresses and makes contributions to a large range of questions, in particular the understanding of symbiotic autonomy that enables the robots to proactively ask for help to overcome their limitations.
“But today, reaching the 1,000 km landmark is a celebration of long-term robot localization and navigation, and the remarkable novel underlying algorithms that are the core of the PhD thesis of my student, Joydeep Biswas,” Veloso said. “Without such work, our robots would simply not move autonomously.”
The three CoBots — machines that look something like mobile podiums — have omnidirectional wheels that enable them to move in any direction and sensors that allow them to detect and avoid hitting people and other obstacles that might not be on their onboard maps.
The CoBots operate without any researcher following them around, and traverse all the wide, varied spaces in the CMU buildings, ranging from multi-width and multi-angled corridors and hallways to glass bridges, open spaces with unique furniture and crowded café areas.
“The scale of the deployment also opened the novel challenge of collecting and understanding data about the extensive robots’ performance, as we clearly do not follow the robots to observe them,” said Biswas, who is pursuing a Ph.D. in robotics.
Each robot gathers daily logs of their errands in the buildings, which are then processed to report on the performance of the algorithms, as well as to learn patterns that improve the robot maps of the buildings. Biswas receives emails from the robots each night about the day’s activities. With three CoBots in routine operation, that means a lot of data about the robots and their environment gets collected and the researchers have had to develop new methods for handling and mining all of it.
The 1,000 km milestone for the CoBots is a major achievement also due to the navigation challenges of indoor mobile robots. A building does not include reliable global location information, such as the GPS signals available outdoors, and includes repetitiveness that creates ambiguity, and open spaces without static features or with dynamic changing features, as people move around.
“We devised a new localization algorithm capable of handling static features, such as walls, and short-term features whose positions vary, such as tables and chairs, while distinguishing them from dynamic features, such as moving people,” Biswas said.
CoBots are robustly safe in their close interaction with people.
“We have operated without a single incident,” Veloso said. “They never went over a person’s foot, never ran into a wall, never got confused by a glass window.
“We wish to see the CoBots around us for many more additional 1,000km!” she added.