We've seen robot dogs run up hills with luggage, and help fight fires. Now, researchers at Switzerland's ETH Zurich are putting these mechanical mutts through their paces on the badminton court, teaching them to play about as well as a seven-year-old human.
The engineers' new control system for the institute's four-legged ANYmal robot uses two cameras to track the shuttlecock as it moves through the air, predicts its flight trajectory, and then navigates the quadruped to the optimal position in its space to return the shot using a racket mounted to a multi-axis arm. Watch ANYmal in action in the clip below:
What's neat about this is the reinforcement learning program informs the control system for movement and using the racket arm. It essentially simulates countless shots in a virtual environment to give ANYmal a better sense of how to return a shuttlecock coming its way.
This enables the bot to accurately intercept the shuttlecock, while also moving quickly and without losing its balance. That requires plenty of coordination between its visual perception systems and motor functions.
While the reinforcement learning-based control system is new, ANYmal has been around for several years now, and is commercially available for autonomous inspection routines in complex environments like detecting gas leaks in power plants. Its movement capabilities have been massively upgraded recently, so it's properly agile.
It had previously learned to navigate bumpy hiking trails, and learned parkour last year. That unlocked its ability to balance itself on uneven terrain, making it a potentially useful companion in search and rescue missions where disasters have reduced buildings to rubble.
With this latest development, ANYmal could be deployed in a range of sports training assistance roles. The bot is said to cost about US$150,000 though, so I'd expect to only see this show up in elite teams' facilities when the tech is ready.
Source: ETH Zurich
