NTT recently announced further plans related to their physical AI ambitions by developing a system which lets robots assess how cargo will respond to their actions to help prevent damage.
The evaluation technology is expected to contribute to physical AI ambitions in various industries, though NTT is plotting an initial foray in the realm of airport luggage handling.
NTT stated it aims to provide information to prevent crushing or toppling when robots lift loads.
It developed a proprietary physical control adapter to assess corporeal elements of objects and predict how they might change when machines interact. A deep-learning image generation model can then visualise likely reactions.
NTT explained current approaches suffer from inaccuracies and a lack of awareness which could result in collisions. Perhaps more pressing, “they rely on large amounts of training data”, which can limit performance when working beyond learned scenarios.
The NTT World Model can be integrated with existing robot control systems, adding a prediction layer and using the physical characteristics of the items being lifted to go beyond the limits of training data.
It reported a near 20% improvement in success rates during tests where a small robot simulated cargo stacking.
NTT listed logistics and warehousing, manufacturing, construction and retail as potential fields the technology could be applied to.
It intends to work with robotics software provider Mujin on future demonstrations and so lay the groundwork for physical AI.
Source: Mobile World Live
Image Credit: ShutterStock
Source: Tahawul Tech