Researchers in the United States have developed a new learning framework for humanoid robots that allows them to better collaborate with humans, whose movements are often unpredictable. This approach moves away from treating humans as a predictable factor and instead focuses on improving the robot's flexibility in response to unexpected human gestures. Companies such as Figure AI, which are designing humanoid robots, are exploring applications in both industrial and home environments. Previously, simulations trained robots by assuming humans would act predictably, but this often led to failures when human behavior changed in terms of movement, rhythm, or decisions.
A team from Carnegie Mellon University has created a framework called Heterogeneous-Agent Lyapunov policy Optimization (HALO), designed to address the lack of flexibility in machines' responses to spontaneous human behavior. This framework uses multi-agent reinforcement learning, allowing robots to autonomously learn how to interact and collaborate with humans. In practice, the robot does not follow a fixed routine but is placed in an environment where its human partner can display a wide range of behaviors. This means the robot incorporates human variability directly into its learning process.
In collaborative robotics, two agents—such as a robot and a human—can have the same objective, like moving furniture or other objects in a room, but may use different strategies. This can lead to a "rational gap," where both agents are unsure of how to proceed. To stabilize this decentralized learning process, the HALO method uses Lyapunov stability theory, a mathematical concept that ensures, despite initial differences in movement or rhythm, the robot and human can smoothly and safely reach a shared goal. Researchers have shown that using HALO, robots can navigate around furniture and walls while carrying heavy and awkward objects. They can quickly adjust their posture or movement to better handle changes in weight or unexpected human actions, ensuring a more effective collaboration.
The HALO method has already been tested in real-world conditions and is expected to have applications in several areas. The industrial and logistics sectors could benefit significantly, especially in the transport of heavy loads and complex assembly tasks. Scientists also believe the method could improve the efficiency of machines assisting human rescuers in time-sensitive situations. Additionally, the researchers have mentioned the potential use of HALO in hospital environments, where it could help with the smooth and safe transport and handling of patients, such as during bed stretcher transfers or other medical procedures, without causing jolts or discomfort.
New Learning Framework Enhances Humanoid Robot Collaboration with Humans
AI-rewritten from original reportingHow it works
roboticsaicollaborationhalocarnegie-mellonreinforcement-learning



