Autonomous robot navigates around pedestrians
Image credit: MIT
Engineers at Massachusetts Institute of Technology (MIT) have developed a robot capable of moving carefully around pedestrian-heavy areas such as pavements, shopping centres and hospitals.
While great pains have been taken to integrate the rules of the roads into self-driving cars for the sake of safety, little attention has been given to teaching robots how to behave around foot traffic.
On the occasions when robots have taken to pedestrian-only areas, such as pavements, they have unsurprisingly failed to integrate with foot traffic, causing havoc.
For humans – and perhaps our pet dogs – it is instinctive to weave between other pedestrians, avoid obstacles and walk at a steady pace. Programming this “socially aware navigation” into autonomous robots, however, is a challenge.
“Socially aware navigation is a central capability for mobile robots operating in environments that require frequent interactions with pedestrians,” said Yu Fan “Steven” Chen, who led the study as a graduate student at MIT.
“For instance, small robots could operate on sidewalks for package and food delivery. Similarly, personal mobility devices could transport people in large, crowded spaces, such as shopping malls, airports and hospitals.”
The researchers began by identifying four main issues that a robot would have to tackle to navigate through an area with heavy foot traffic: localisation (identifying its location), perception, control and motion planning.
Three of these four problems can be solved with standard solutions: a robot can identify its location using existing algorithms, observe its surroundings with webcams and other sensors, and be controlled with methods developed to drive autonomous cars. Motion planning remained the only major challenge in creating pedestrian-friendly autonomous robots.
“Once you figure out where you are in the world, and know how to follow trajectories, which trajectories should you be following?” said Michael Everett, another MIT graduate student.
Rather than program the robot to find a single optimal path based on assumed trajectories – which would be slow and unreliable – or to program it to avoid collisions, the researchers followed a new machine learning approach.
“Reinforcement learning” uses computer simulations to train the robot to navigate around virtual pedestrians. Over the course of the training, the robot learns to consider its environment and other objects as it chooses its path, and re-assesses its path ten times a second.
The researchers also incorporated social norms, such as encouraging the robot to pass on the right.
“We want it to be travelling naturally among people and not be intrusive,” said Everett. “We want it to be following the same rules as everyone else.”
The engineers performed test drives in the busy corridors of MIT’s Stata Building. The robot, which is the height of a toddler, was able to successfully navigate through a pedestrian-heavy area avoiding collisions for 20 minutes.
“We wanted to bring it somewhere where people were doing their everyday things, going to class, getting food, and we showed we were pretty robust to all that,” said Everett. “One time there was even a tour group, and it perfectly avoided them.”