Areas of Research

Today's AI models struggle to transfer learned skills to new physical tasks, especially those requiring precise force control and tight tolerances. RAI is developing foundation models that allow robots to accomplish complex physical tasks by combining novel data-driven learning techniques with principled models that exploit the full capabilities of a robotic system.

RAI is developing the methods, algorithms, data, and hardware needed for robots to dynamically manipulate physical objects and systems to perform useful tasks, such as assembly, repair and transport. Our research is aimed at going beyond the static grasping and manipulation that currently dominates the field.

To achieve natural, robust dynamic behavior, robots must safely interact with their environment and each other. RAI is tackling this through reinforcement learning for control, allowing high-level direction by AI algorithms, incremental learning of new behaviors, and adaptation to novel environments.

Building advanced sensors that provide robots (including wheeled, humanoid, and quadruped designs) with a deep understanding of their surroundings. This capability, combined with our robust mobility solutions, ensures they can effortlessly navigate and traverse a wide variety of environments, from indoor spaces to the most challenging terrain.

Integrating robots into daily life will have an impact on human society. Our approach combines rigorous scientific research and data collection with a deep understanding of the socio-technical systems — including the social, legal, and ethical contexts — that shape robot integration and human attitudes towards it.