InEKFormer: A Hybrid State Estimator for Humanoid Robots

TL;DR

This paper tackles the challenge of accurate floating-base state estimation for humanoid locomotion under complex contact transitions. It introduces InEKFormer, a hybrid estimator that internally integrates the invariant extended Kalman filter ($InEKF$) with a Transformer to predict Kalman gains, using a fixed $SE_4(3)$-state augmented with foot contacts. The authors provide a new RH5 humanoid dataset with ground-truth motion capture and proprioceptive data, and comprehensively compare InEKFormer to InEKF and KalmanNet, showing Transformer-based gains can offer superior performance when trained autoregressively. The work highlights the potential of Transformer-based gain estimation for high-dimensional humanoid systems and points to future improvements in autoregressive training, sim-to-real transfer, and broader generalization to other robot platforms.

Abstract

Humanoid robots have great potential for a wide range of applications, including industrial and domestic use, healthcare, and search and rescue missions. However, bipedal locomotion in different environments is still a challenge when it comes to performing stable and dynamic movements. This is where state estimation plays a crucial role, providing fast and accurate feedback of the robot's floating base state to the motion controller. Although classical state estimation methods such as Kalman filters are widely used in robotics, they require expert knowledge to fine-tune the noise parameters. Due to recent advances in the field of machine learning, deep learning methods are increasingly used for state estimation tasks. In this work, we propose the InEKFormer, a novel hybrid state estimation method that incorporates an invariant extended Kalman filter (InEKF) and a Transformer network. We compare our method with the InEKF and the KalmanNet approaches on datasets obtained from the humanoid robot RH5. The results indicate the potential of Transformers in humanoid state estimation, but also highlight the need for robust autoregressive training in these high-dimensional problems.

Figures (7)

• Figure 1: Hybrid state estimation approach consisting of a model-based InEKF and a data-driven Transformer network for the RH5 humanoid robot.
• Figure 2: Block diagram of the high-level InEKFormer architecture, where $\frac{1}{\mathrm{z}^k}$ denotes $k$ unit delays.
• Figure 3: InEKFormer gain estimator model architecture.
• Figure 4: RH5 robot coordinate frames and sensing hardware: 1 - IMU, 2 - Linear joint encoders, 3 - FT sensors.
• Figure 5: Overview of the RH5 dataset.