Robot Acquires New Skill: Identifying Object Properties from a Distance

Scientists at a research team have developed a technique called "Micro Motion Inference" that enables robots to infer the material properties of objects, such as their mass or softness, through simple handling movements. The team utilized the open-source Warp library to build this inference system, which supports micro-motion analysis. This method requires only one observation of the robot moving an object along a real trajectory to complete its calculations within seconds. "From a technical perspective, as long as you understand the physical model of the object and the force application method of the robot, you can derive the target parameters," says Chao Liu, a researcher involved in the project. Currently, the experiments focus on identifying the mass and softness of objects. However, this method can be equally applicable for measuring other characteristics, such as friction or fluid viscosity, within enclosed containers. Unlike traditional approaches that rely heavily on computer vision or external sensors, this technique does not require a large dataset for training. It can function effectively in unknown environments or with novel objects, demonstrating greater robustness. "This research is not designed to replace computer vision; both methods have their advantages. But we've shown that even without a camera, some properties can still be accurately inferred," explains Yichen Chen, another member of the team. They also plan to extend the study to more complex systems like soft robotics and fluid dynamics, including interactions with substances like sand and liquids. In the long term, the researchers aim to enhance a robot's learning capabilities, enabling it to quickly grasp new manipulation skills and adapt to environmental changes. Miles Macklin, a high-level supervisor of the research at NVIDIA, comments, "Inferring the physical properties of an object solely from limited or noisy data has always been a challenging problem in robotics. This research demonstrates that robots need only internal force sensors to accurately diagnose properties like mass and softness, eliminating the need for external cameras or specialized measurement tools." The original research can be found at: https://news.mit.edu/2025/system-lets-robots-identify-objects-properties-through-handling-0508 This innovative approach holds significant promise for future advancements in robotics, particularly in scenarios where precise and adaptive manipulation is crucial. By integrating internal force sensors with sophisticated algorithms, robots can now perform tasks more reliably and efficiently, even in unfamiliar settings.