Produced by: Tarun Mishra Designed by: Manoj Kumar
A team of AI specialists at Google's DeepMind have unveiled their groundbreaking achievement in training tiny robots to play soccer, detailed in journal Science Robotics.
Unlike traditional methods, the research team applied machine learning to create remarkably skilled tiny robots, approximately 510 mm tall, capable of playing soccer with impressive proficiency.
The robots are trained using reinforcement learning techniques in computer simulations, mastering skills such as getting up after falling and attempting to score goals. They are then trained to play full one-on-one soccer matches using vast amounts of video and data.
Once the virtual robots demonstrate proficiency, the system is transferred to Robotis OP3 robots, enabling them to learn and improve their skills through testing and gameplay on a small soccer field.
Observations revealed that the AI-trained robots executed movements more smoothly and elegantly compared to those trained using conventional methods, demonstrating faster recovery after falling and employing strategic techniques to outmanoeuvre opponents.
The robots employ advanced strategies, such as feigning turns to deceive opponents and create scoring opportunities, showcasing their superior gameplay compared to robots trained using other techniques.
The successful development of AI-trained robots with remarkable soccer skills opens doors to a wide range of applications, highlighting the potential of machine learning in advancing robotics and automation.