As UC Berkeley Ph.D. student Jun Zeng puts on a blindfold, he is led by a guide dog through a maze of cardboard boxes. They reach a narrow path, and the guide dog pauses, readjusts, then pulls on the leash to let Zeng know it is safe to continue.
But this is no ordinary guide dog. This is a robot.
In a paper published March 26, campus researchers modified a quadrupedal robot from the Massachusetts Institute of Technology into a robot guide dog.
Zhongyu Li is a lead Ph.D. student on a team from the lab led by Koushil Sreenath, campus assistant professor of mechanical engineering. Li said robotic guide dogs will save a lot of time and energy that goes into training a real guide dog.
“Training a dog is very intense, we need to train a dog specifically to teach it how to serve as a guide dog. These leading skills cannot be transferred to another animal,” Li said. “By using a robot dog, we can easily transfer to another machine. We can create multiple body guide dogs to help all kinds of individual people.”
The robot is equipped with a rotating camera to keep track of the person being led and light detection and ranging, or LIDAR, to locate and position itself.
But what sets this robot guide dog apart from others is its leash and its four legs, which help it guide users through narrow spaces such as doors, narrow corridors and corners, according to the paper.
When the robot needs to stop and readjust, Li said it slackens the leash. Once repositioned, the robot will pull the leash taut to let the user know it is safe to continue. The paper added that the leash system replaces previous methods of using a rigid arm as a guiding cane, which can get stuck in tight spaces.
The paper also noted that using legs instead of wheels reduces the space the robot takes up.
“Most prior robotic guiding systems are based on wheeled platforms with large bases,” the paper reads. “The large bases … limit these prior approaches from operating in narrow and cluttered environments.”
According to the paper, the robot guide dog successfully led its users to the end goal without any collisions.
Li said the team has only done indoor research so far but hopes to expand research into outdoor environments, such as leading someone to cross a traffic light. The paper noted that future research would also look into more complicated human movements.
Li added that he is hopeful that this technology will hit the market very soon.
“People can get a robot dog for a very cheap price in the future soon,” Li said. “Once we develop one algorithm, we can copy-paste for all kinds of robot dogs. It won’t be too long. The price is getting lower and lower nowadays.”
