Since Boston Dynamics debuted its famous legged robot in 2013, many organizations have unveiled machines capable of autonomously performing a variety of professional and recreational activities. Latest research uses badminton to improve robot agility and coordination, which could help improve training methods and mobility.
Scientists at ETH Zurich recently published a study and video (below) explaining how they trained a four-legged robot to play badminton. This study could demonstrate the effectiveness of a full-body training model that does not break down activities into separate steps, potentially expanding the range of learnable tasks.
Standard ANYmal-D robots are equipped with DynaArm and trained using standard reinforcement learning models without excessive software or hardware specialization. Additionally, the machine learned to track shuttlecocks using only its built-in stereo camera.
The robot learned to navigate the court using NVIDIA's Isaac Gym virtual simulator, performing approximately 7,500 iterations over nearly five hours using an RTX 2080 Ti graphics card. In addition, the noise prediction model can help it track the shuttlecock when it flies out of the camera's limited field of view.
Eventually, it learned to hit a badminton ball using a variety of complex movements, such as starting, backswing, hitting, running and recovering, without the need for researchers to guide them one by one. It even adjusts its movements based on the perceived speed of the shuttlecock, racing around the court to hit the ball in time.
This experiment demonstrates the adaptability of ANYmal-D. ANYmal-D is designed by ANYbotics to perform inspections, emergency response and other tasks in a variety of industrial environments. The company's website shows the robot operating around the clock and navigating indoor facilities, gravel roads, stairs, caves, snow, mud and flooded environments. The upcoming ANYmal-X model is certified for use in more hazardous environments to help perform inspections involving oil, gas and chemicals.
Badminton is just the latest recreational activity to demonstrate the adaptability of legged robots. Boston Dynamics has shown off its original Atlas model and demonstrated dance and gymnastics performances, while the company's four-legged robot Spot has demonstrated how robots can assist stage productions in the future.
Meanwhile, Google DeepMind trained a robotic arm to play table tennis, Engineered Arts showcased a humanoid robot artist, and the NBA recently deployed robots to assist athlete training. Chinese companies have also recently deployed robots in half-marathons and kickboxing competitions, with mixed results.