NTT Develops Physical AI System to Help Robots Safely Handle Cargo

Image: Illustrative image · ESO · CC BY 4.0 · Source

NTT has created a physical AI system enabling robots to anticipate how cargo reacts to their actions, aiming to reduce damage in tasks like airport luggage handling.

Japanese technology firm NTT has advanced its physical AI initiatives with the development of a new system designed to help robots better handle cargo by predicting how objects will respond to manipulation. This technology aims to prevent damage such as crushing or toppling when robots lift loads, a common concern in automated logistics and cargo management.

The system uses a proprietary physical control adapter that assesses the tangible properties of objects, providing data on how these items might deform or move when interacted with. Leveraging a deep-learning image generation model, the technology can visualize potential outcomes of physical interactions, offering robots a predictive layer to guide their movements.

Traditional approaches to robotic manipulation often struggle with inaccuracies and a limited understanding of object physicality, sometimes requiring extensive training data tied to specific scenarios. NTT's innovation addresses these challenges by incorporating real-time physical characteristics, allowing the system to operate effectively even beyond previously learned conditions.

The "NTT World Model" integrates with existing robot control setups, enhancing their ability to foresee and adapt to physical changes during handling tasks. In trials involving a small robot simulating the stacking of cargo, the new system improved success rates by nearly 20%.

While initial focus is on airport luggage handling, NTT envisions applications across industries including logistics, warehousing, manufacturing, construction, and retail. The company plans to collaborate with robotics software provider Mujin on upcoming demonstrations, supporting broader adoption of physical AI technologies.

This development reflects growing efforts to improve robotic autonomy and safety in physical environments, potentially transforming operational efficiency and asset protection in multiple sectors.