Robotic, AI Guide Dog Under Development

by Financial Review

   Guide dogs are a great help to vision-impaired and blind people, assisting them to navigate their surroundings and get to where they need to go.

   But they’re not for everyone. Some people have allergies that make dogs unsuitable; others might not want to take on the responsibility of looking after them.

   Developing a replacement is a solution to the problem being worked on by Sarath Kodagoda, a researcher in robotics and mechatronics at the University of Technology Sydney (UTS).

   Professor Kodagoda and his colleagues are developing a robotic guide dog that enhances mobility, independence and social participation for the blind community.

   UTS was a finalist in the Community Engagement category of the inaugural Financial Review AI Awards for its work on the robotic dog.

   The robot itself takes the form of a dog and is built on top of an off-the-shelf Unitree Go2 robotic dog.

   “The main reason for choosing that is they are very, very agile. They can climb stairs. They can walk on various terrains. They don’t need that to be flat,” Kodagoda says.

   “And the other reason is that we don’t want to reinvent the wheel. If something is available, we use that and then build the rest of the intelligence and the requirements from there.”

Laser range finder

   The robotic guide dog is driven by AI and real-time decision-making capabilities to guide individuals based on their own requirements. The system interprets user intentions, adjusts to environmental changes, and provides auditory and haptic feedback – such as vibration – to the users.

   Kodagoda says the ambition is to create a system that can describe where users are, what’s around them and answer any questions they might have.

   “If I say I want to go somewhere such as to a coffee shop, if it finds out where the close-by coffee shops are and then talks to this person and says, ‘would you like to go to this place?’ and then interactively, once it is decided, plan the path, guide that person very safely,” he says.

   “When you enter the coffee shop, it might say that on your right-hand side, there are four empty chairs and a table, and then talk about the menu and help them select.”

   The robotic dogs will use a range of sensors, possibly including a laser range finder or LiDAR, which is the same technology that autonomous vehicles use.

   However, Kodagoda says that building a robotic guide dog is in many ways more complicated that building a self-driving car.

   While cars obviously move at much greater speeds, they are mostly moving with other cars going in the same direction. The robotic guide dogs, by contrast, will have to guide their users through areas such as train stations or shopping centers where there are potentially many people going in many different directions.

   They also have to learn what Kodagoda calls “socially acceptable navigation.”

   “For example, in the office environment, if two people are talking, we don’t go in between them,” he says. “So this will learn the behaviors of a particular environment and then guide that person according to the general social norms.”

   The researchers have held numerous workshops with members of the blind community to get an understanding of their needs and to test the technology as they develop it.

Scaled-down version

   Kodagoda says one exercise that was very useful was when the researchers were blindfolded and guided around. They learned the importance, for instance, of being prepared to pass through a narrow space. If the robotic dog is taking users through a door, it warns them that a door is approaching so they are prepared.

   While the system is being built using robotic dogs, Kodagoda says there is no intention to replace actual guide dogs for most users and it could be used in a format other than the dog platform.

   The researchers are currently working on a minimum viable product and are designing it in a modular way, so different parts can be used in different applications. For instance, a user could wear a scaled-down version in a harness.

   The modular design will also allow the solution to evolve and improve as new technologies become available.

   The project came about after Guide Dogs approached the UTS Robotics Institute, and the two organizations have worked together and won additional project funding from the TPG Telecom Foundation, Guide Dogs and a major Australian Research Council Linkage project grant.

   Kodagoda says that ultimately the researchers and Guide Dogs Australia would like to commercialize the technology so it can be made available to more people and have a significant impact.

   The robotic guide dog is one of Kodagoda’s favorite projects.

   “I’ve been working in robotics for 27 years now. This is one of the great projects that I think can address a significant problem and be very impactful,” he says.