UniSC students have taught robodogs new tricks as part of a research project to develop an app that uses voice commands to herd sheep. 

A previous student project had trained Unitree Go2 robodogs to herd sheep autonomously, using cameras, sensors and artificial intelligence to detect livestock and navigate around them. 

The wider research project began after UniSC staff were approached about using the robodogs to support farmers by helping with livestock handling, which can be costly, labour-intensive and physically demanding.

Since then, students have worked alongside farmers and working-dog handlers to explore how robotics could be used to assist livestock operations by improving safety and reducing pressure on both people and animals.

Although the robodogs were originally intended to operate autonomously, working-dog handlers gave feedback indicating they wanted to remain in control of the technology rather than rely solely on automation, inspiring the new branch of the project.  

Project supervisor Dr Erica Mealy said a big part of advancing the research was listening to the people who would use the technology. 

"This project came directly from feedback from working dog handlers,” Dr Mealy said.  

“We wanted the robodog to stop when a handler says stop or move around the back of the flock when a handler gave a command.”

Honours student Abdullah al Muhsin said the voice-command project centred on developing a web-based app that allows users to issue commands via a phone or laptop. 

"The app listens to what you're saying, recognises specific commands and sends those instructions to the robodog,” he said.

The system uses Google Chrome's built-in speech-recognition technology to listen for basic commands such as "stop", "away" and "come by". 

When a user speaks, the app converts their words into text and identifies any recognised commands, transmitting the instruction to the robodog via a WebSocket API, creating a live link between the browser and the robodog. 

Software running on an onboard NVIDIA Jetson computer receives the command and translates it into movement, allowing the robodog to respond in real time. 

The robodog combines those instructions with data from its cameras and sensors, allowing it to detect sheep, maintain a safe distance from the flock and avoid collisions with animals or obstacles. 

Abdullah said the software was designed to work with any user rather than being trained to recognise a specific voice, which was made possible by using Google Chrome's speech-recognition technology. 

"The system isn't trained to a specific person's voice – anybody can use it,” 

The previous robodog project was featured at the Ekka in the Main Arena last year, where the technology was demonstrated alongside sheepdog handlers in front of crowds of about 30,000 people per show.  

The project will return to centre stage at this year's Ekka, celebrating 150 years of the Royal Easter Show, with a nightly demonstration of how the technology can work alongside dogs to herd and control sheep.  

Catch them in the Main Arena every night from 6.30pm as part of EkkaNITES.