State Estimation for 2-Legged Robots Using Foot Slippage and Body Impact Detection

Conventional state estimation methods for legged robots assume stable foot contact and rely on force sensors. However, in dynamic locomotion, these assumptions often break down due to foot slippage and body impacts, leading to significant estimation errors. This paper proposes a probabilistic state estimation framework that operates without contact sensors, integrating foot contact inference, slip detection, and body impact estimation into a unified model. Contact state estimation is performed using a momentum-based disturbance force estimation method, while slip state estimation distinguishes between sliding and stick-slip. Additionally, body impact states are probabilistically estimated based on angular velocity, linear acceleration, and body tilt information. The proposed framework is validated through simulations under various ground friction conditions and step times. Compared to conventional contact-based estimation methods, the proposed method reduces position estimation errors by 81.8% in the plastic foot friction model and 75.9% in the rubber foot friction model. Furthermore, the velocity errors are reduced by 51.3% and 47.1% in plastic and rubber surface conditions, respectively.