Active Embodiment Identification with Reinforcement Learning for Legged Robots
This work addresses the problem of autonomous embodiment identification for legged robots, which is crucial for adaptation and control, but the results are demonstrated only in simulation.
The paper presents a reinforcement learning method for legged robots to actively identify their own embodiment (joint-level and global parameters) through interaction, achieving accurate inference across different morphologies in simulation.
We present an active embodiment identification method for legged robots that jointly learns information-seeking behavior and explicit embodiment prediction. Using a history-augmented URMA architecture, the method infers joint-level and global embodiment parameters through interaction with the environment in simulation across different morphologies.