Although legged animals are capable of performing explosive motions while traversing confined spaces, replicating this behavior in quadrupedal robots has been a longstanding challenge. Here, we propose a hierarchical reinforcement learning pipeline that empowers the robots to perform aggressive locomotion through constrained obstacles--a narrow gate. The imitation learning technique is used to train the low-level policy, which mimics the behaviors of real animals and forms a set of diverse skills. The high-level controller, having an awareness of the capability of low-level skills and acquiring the gate information via vision-based detection, determines the suitable maneuvers with collision-free trajectories to traverse it dynamically. Notably, we also verify that this framework can be extended to other highly dynamic tasks. This is one of the first works that perform autonomous and agile aerial gate traversal tasks on ground-walking robots, extending the lifelike agility of legged robots to match that of their biological counterparts.
Zeren Luo, Jiahui Zhang, Yimin Han et al.· 1 citation
Rice is one of the most significant crops in all the world agriculture, diseases and pests of rice seriously threaten grain production. Identifying and classifying them accurately in the field is essential for prevention. However, the existing models often encounter issues such as unequal between lightweight deployment and recognition performance, weak robustness to various lighting conditions and insufficient generalization ability due to class imbalance. To address this issue, the paper proposed a lightweight method for rice diseases' classification based on the optimized MobileNetV2. Its core structure adopted MobileNetV2 as the backbone to retain the advantage of depthwise separable convolution lightweight model and added an SE attention layer after feature extraction stage to enhance the model's focus on lesion regions. Using Test-time- augmentation (TTA) further improves the model's robustness to lighting and shooting angle interference. Based on the experiment results, the proposed model achieves a test average accuracy of 83.9% and a maximum validation accuracy of 88.0% on the testing sets. Under extreme lighting conditions, the performance loss percentage is controlled at 20.78%. The rice disease identification method proposed in this paper provides a reference for the deployment and application of the method in different light conditions in the field.
