In this paper, we propose a biped walking controller that optimized three push recovery strategies: the ankle, hip, and stepping strategies. We suggested formulations that related the effects of each strategy to the stability of walking based on the linear inverted pendulum with flywheel model. With these relations, we could set up an optimization problem that integrates all the strategies, including step time change. These strategies are not applied hierarchically, but applied according to each weighting factor. Various combinations of weighting factors can be used to determine how the robot should respond to an external push. The optimization problem derived here includes many nonlinear components, but it has been linearized though some assumptions and it can be applied to a robot in real time. Our method is designed to be robust to modeling errors or weak perturbations, by exploiting the advantages of the foot. Hence, it is very practical to apply this algorithm to a real robot. The effectiveness of the walking controller has been verified through abstracted model simulation, full dynamics simulation, and a practical robot experiments.

This paper proposes a deep learning-based anomaly detection method using time-series vibration and current data, which were obtained from endurance tests on driving modules applied in industrial robots and machine systems. Unlike traditional classification models that depend on labeled fault data for detection, acquiring sufficient fault data in real industrial environments is highly challenging due to various conditions and constraints. To address this issue, we employ a semi-supervised learning approach that relies solely on normal data to effectively detect abnormal patterns, overcoming the limitations of conventional methods. The performance of semi-supervised models was first validated using a statistical feature-based anomaly detection approach, from which the GCN-VAE model was adopted. By combining the spatial feature extraction capability of Graph Convolutional Networks (GCNs) with the latent temporal feature modeling of Variational Autoencoders (VAEs), our method can effectively detect abnormal signs in the data, particularly in the lead-up to system failures. The experimental results confirmed that the proposed GCN-VAE model outperformed existing hybrid deep learning models in terms of anomaly detection performance in the pre-failure section.

This paper presents an efficient approach to collision detection for real time controlled humanoid robot. The proposed method is based on separating axis theorem (SAT) for oriented bounded box (OBB). We construct tight-fitting OBB based on 3D-CAD model of the humanoid robot and OBB trees for efficient computing. In addition, in order to solve several potential problems when detecting collisions at discrete time steps, we define a velocity sphere and operate Minkowski sum of the OBB and its velocity sphere. The collision detection algorithm is applied to real humanoid robot, DRC-HUBO+ and many experimental results demonstrate the feasibility of the proposed approach.