Active Embodiment Identification with Reinforcement Learning for Legged Robots

TL;DR

This paper introduces a reinforcement learning-based method for legged robots to actively identify their own embodiment by learning to seek information and predict their physical parameters through interaction.

Contribution

It proposes a novel history-augmented URMA architecture that jointly learns embodiment prediction and information-seeking behaviors in simulation.

Findings

Successfully infers joint-level and global embodiment parameters.

Demonstrates effective embodiment identification across different morphologies.

Enhances robot adaptability through active self-awareness.

Abstract

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.