A research team from Wuhan University in China has announced the development of an advanced artificial intelligence framework called RGMP, marking a significant leap in the realm of domestic robots and promising a transformative shift in their capabilities. This innovative system empowers humanoid robots with enhanced understanding and precise execution of household tasks.
The RGMP system relies on a design that drastically reduces the need for extensive training data, paving the way for robots that learn quickly and efficiently. It is based on a sophisticated blend of machine learning and geometric reasoning, enabling the robot to analyze the shape of the object it is interacting with and determine the most appropriate action, whether it involves picking, pushing, or gripping.
Unlike traditional systems that rely heavily on large datasets, RGMP leverages this integration to equip the robot with the ability to handle new and unfamiliar environments without extensive training. The system comprises two key components:
The first is the Geometric Skill Selector (GSS), which helps the robot choose the appropriate type of movement in a way that mirrors human decision-making. The second is the Adaptive Relational Graph Network (ARGN), which allows the robot to learn rapidly from a very limited number of examples by storing and updating spatial memory during actual interaction.
In a series of tests, researchers applied the system to a humanoid robot and a bimanual robot equipped with cameras, relying on only 120 demonstration trials. Despite this limited number, RGMP outperformed leading global models in the field, such as Diffusion Policy, OpenVLA, and ResNet50, recording a remarkable increase in robot performance. The system achieved a 25% increase in skill selection accuracy and demonstrated a high ability to execute complex movements with greater stability. Notably, it achieved strong results using only 40 training examples, compared to the approximately 200 examples required by traditional systems.
The research team believes that integrating symbolic reasoning with deep learning represents a pivotal step towards a new generation of more autonomous and intelligent robots. Researchers are currently working on developing a future version of RGMP that could allow the robot to learn a new task from watching just one example, paving the way for the creation of robots capable of instantly adapting to daily changes and handling household tasks with unprecedented flexibility.
