The systems are far from being good. Although the desk tennis robot excelled against novice human opponents, beating all of them and a significant 55% of participants at this level, it faltered against advanced players, losing all matches. Nonetheless, it’s a powerful advance.
For several months now, our projections have consistently indicated that the robotic would struggle to prevail against individuals with whom it had no prior experience. Pannag Sanketi, a seasoned senior software program engineer at Google DeepMind, remarks that the system far surpassed their initial expectations. “The robotic system’s greatest asset was its ability to quickly and effectively process complex information.”
The analysis isn’t limited to just entertaining and video games. Indeed, this development marks a significant stride towards constructing robots capable of performing useful tasks such as property maintenance and warehouse management. According to Lerrel Pinto, a computer science researcher at New York University not involved in the project, Google DeepMind’s approach to coaching machines has far-reaching implications for multiple sectors.
As he enthusiastically remarks, “I’m a huge proponent of witnessing robotic systems seamlessly integrate with real people – it’s an exceptional example of this happening.” Despite its limitations, the raw ingredients exist to continually improve and ultimately reach their potential.
To become a skilled table tennis player, one needs exceptional hand-eye coordination, rapid agility, and the ability to make swift decisions in response to an opponent’s moves – all attributes that pose significant challenges for robots. Researchers at Google DeepMind developed a dual-approach methodology to instruct the AI system to mimic these skills: first, they employed laptop simulations to train the system in mastering its batting abilities, followed by fine-tuning using real-world data that enables continuous improvement.
