2 June 2025 in IEEE Sensors Journal
This study aims to enhance the accuracy of a six-axis force/torque (F/T) sensor by improving upon existing approaches that utilize a multilayer perceptron (MLP) and the least-squares method. While previous research has employed MLPs for thermal compensation, it has not effectively addressed temperatureinduced drift. The sensor used in this study operates based on infrared light and incorporates a photocoupler, which makes it highly sensitive to dark current effects, resulting in significant drift under temperature variations. Moreover, its compact and lightweight design (45 g) leads to low thermal capacity, making it susceptible to rapid temperature fluctuations even with minimal heat input, which in turn affects real-time performance. To address these challenges, this study proposes a gated recurrent unit (GRU)-based method and compares it with the conventional MLP approach. Experimental results demonstrate that the GRU-based model significantly reduces drift from 26 N to 2.7 N, and the root mean square error (RMSE) from 500 to 3-representing a 100-fold improvement. These findings suggest that GRU-based modeling substantially improves real-time F/T measurements in temperature-sensitive environments, benefiting applications in robotics and precision instrumentation.
12 February 2026 in IEEE Robotics and Automation Letters
In this letter, we propose an Attention-Based Neural-Augmented Kalman Filter (AttenNKF) for state estimation in legged robots. Foot slip is a major source of estimation error: when slip occurs, kinematic measurements violate the no-slip assumption and inject bias during the update step. Our objective is to estimate this slip-induced error and compensate for it. To this end, we augment an Invariant Extended Kalman Filter (InEKF) with a neural compensator that uses an attention mechanism to infer error conditioned on foot-slip severity and then applies this estimate as a post-update compensation to the InEKF state (i.e., after the filter update). The compensator is trained in a latent space, which aims to reduce sensitivity to raw input scales and encourages structured slip-conditioned compensations, while preserving the InEKF recursion. Experiments demonstrate improved performance compared to existing legged-robotstate estimators, particularly under slip-prone conditions.