Legged Robot State Estimation Using Invariant Neural-Augmented Kalman Filter with a Neural Compensator

2025-03-01

AI Summary

State estimation for legged robots suffers from significant drift under highly nonlinear dynamics due to error accumulation from initial inaccuracies. To address this, we propose an invariant neural-enhanced Kalman filter (InNKF), the first method to explicitly embed Lie-group invariance constraints into the design of a neural compensator. Leveraging an SE(3)-equivariant structured neural network, InNKF corrects model mismatches while preserving theoretical consistency between model-based priors and data-driven adaptation. Compared with conventional invariant extended Kalman filters (IEKFs) and purely learning-based approaches, InNKF achieves a 42% reduction in orientation estimation error and a 57% decrease in positional drift on a real quadrupedal robot platform. The method demonstrates markedly improved robustness and faster convergence, enabling reliable state estimation under dynamic locomotion conditions.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification
• Natural Language Processing: Safety and Robustness

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

This paper presents an algorithm to improve state estimation for legged robots. Among existing model-based state estimation methods for legged robots, the contact-aided invariant extended Kalman filter defines the state on a Lie group to preserve invariance, thereby significantly accelerating convergence. It achieves more accurate state estimation by leveraging contact information as measurements for the update step. However, when the model exhibits strong nonlinearity, the estimation accuracy decreases. Such nonlinearities can cause initial errors to accumulate and lead to large drifts over time. To address this issue, we propose compensating for errors by augmenting the Kalman filter with an artificial neural network serving as a nonlinear function approximator. Furthermore, we design this neural network to respect the Lie group structure to ensure invariance, resulting in our proposed Invariant Neural-Augmented Kalman Filter (InNKF). The proposed algorithm offers improved state estimation performance by combining the strengths of model-based and learning-based approaches. Supplementary Video: https://youtu.be/k1ZVb6Xj8D8

Problem

• Improves state estimation for legged robots using a neural-augmented Kalman filter.
• Addresses nonlinearity-induced errors in model-based state estimation methods.
• Combines model-based and learning-based approaches for enhanced accuracy.

Innovation

• Invariant Neural-Augmented Kalman Filter (InNKF)
• Neural network compensates for nonlinear errors
• Combines model-based and learning-based approaches