Robust And Efficient Phase Estimation in legged Robots Via Signal Imaging And Deep Neural Networks

Electronics and Communication Engineering

Reliable phase estimation plays a crucial role in enabling smooth, adaptive, and disturbance-resistant locomotion in legged robots. Traditional estimation methods often rely on handcrafted features or analytical gait models, which suffer from reduced performance under noisy sensor inputs, terrain variability, and rapid gait transitions. This research proposes a robust and efficient phase estimation framework that leverages signal imaging techniques and deep neural network architectures to learn discriminative gait representations directly from raw proprioceptive signals. Sensor streams—including joint kinematics, actuator feedback, and inertial data—are encoded into 2D image-based signal maps to capture the underlying spatial–temporal structure of locomotion dynamics. A customized CNN–RNN hybrid model is designed to process these signal images, providing accurate, real-time phase predictions with high resilience to disturbances and domain shifts. Experimental results on simulated and real quadruped platforms demonstrate significant improvements in accuracy, generalization, and computational efficiency compared to baseline estimators. This approach offers a scalable solution for next-generation legged robots, enhancing adaptability, terrain awareness, and locomotion stability

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.