Interacting Multiple Model Proprioceptive Odometry for Legged Robots

Abstract

State estimation for legged robots remains challenging because legged odometry generally suffers from limited observability and therefore depends critically on measurement constraints to suppress drift. When exteroceptive sensors are unreliable or degraded, such constraints are mainly derived from proprioceptive measurements, particularly contact-related leg kinematics information. However, most existing proprioceptive odometry methods rely on an idealized point-contact assumption, which is often violated during real locomotion. Consequently, the effectiveness of proprioceptive constraints may be significantly reduced, resulting in degraded estimation accuracy. To address these limitations, we propose an interacting multiple model (IMM)-based proprioceptive odometry framework for legged robots. By incorporating multiple contact hypotheses within a unified probabilistic framework, the proposed method enables online mode switching and probabilistic fusion under varying contact conditions. Extensive simulations and real-world experiments demonstrate that the proposed method achieves superior pose estimation accuracy over state-of-the-art methods while maintaining comparable computational efficiency.

Figures (10)

• Figure 1: Illustration of rolling and slippery contacts between the robot foot and the ground. As the robot moves along the X axis, the foot exhibits forward rolling motion about the foot-end radius $\mathbf{r}$.
• Figure 2: Overview of the proposed IMM-based proprioceptive state estimation framework. The framework fuses IMU and joint encoders measurements and instantiates two parallel error-state Kalman filters corresponding to rolling and slippery contact models, whose mode probabilities and fused state provide accurate and robust pose. The reference frames are defined as follows: the inertial frame ${W}$ is fixed to Earth, and the body frame ${B}$ aligns with the IMU frame ${I}$. The four feet are labeled as Left Front (LF), Right Front (RF), Left Hind (LH), and Right Hind (RH).
• Figure 3: Some snapshots of the four simulation scenarios are presented, including flat surface, slippery, uneven terrain, and slope environments.
• Figure 4: Overall performance comparison in Straight line trajectory. Left: Estimated trajectories obtained by different methods are compared with the ground truth. Right: The foot velocity during the contact phase is non-zero, which is consistent with the rolling-contact assumption; the darker shaded background indicates the contact phase.
• Figure 5: Estimated trajectories and attitude errors in the slippery scenario.