Search

Point cloud completion has become increasingly popular among generation tasks of 3D point clouds, as it is a challenging yet indispensable problem to recover the complete shape of a 3D object from its partial observation. In this paper, we propose a novel SeedFormer to improve the ability of detail preservation and recovery in point cloud completion. Unlike previous methods based on a global feature vector, we introduce a new shape representation, namely Patch Seeds, which not only captures general structures from partial inputs but also preserves regional information of local patterns. Then, by integrating seed features into the generation process, we can recover faithful details for complete point clouds in a coarse-to-fine manner. Moreover, we devise an Upsample Transformer by extending the transformer structure into basic operations of point generators, which effectively incorporates spatial and semantic relationships between neighboring points. Qualitative and quantitative evaluations demonstrate that our method outperforms state-of-the-art completion networks on several benchmark datasets. Our code is available at https://github.com/hrzhou2/seedformer., Comment: Camera-ready, to be published in ECCV 2022, with supplementary material

Developing a friendly and efficient obstacle avoidance algorithm for mobile robot in dynamic environments is challenging in the scenarios where robot plans its paths without observing other obstacles' intents. Recent works have shown the power of deep reinforcement learning techniques to learn collision-free policies. However, many of them ignore the interactions between obstacles, which may cause unnatural trajectory. Meanwhile, the performance of these methods is not well in simulation environment. Combined with the interactions of obstacles, we propose a mobile robot obstacle avoidance algorithm based on deep reinforcement learning in dynamic environments. Firstly, we not only consider the direct impact of obstacles on robots, but also take the interaction between obstacles into account using angel pedestrian grid. Therefore, robot can extract more abundant environmental prior information. Subsequently, the temporal characteristics of obstacles are extracted through attention mechanism, then we can obtain the joint impact of obstacles on the obstacle avoidance strategy. Finally, the robot obtains the control output through value-based reinforcement learning. The feasibility and effectiveness of our proposed algorithm in dynamic environments is verified through simulation experiments.

In order to improve the utilization of renewable energy and reduce the damage caused by the volatility of renewable energy, it is necessary to further study the multi-energy combined operation strategy. In this paper, a combined operation mode of wind power, gas-fired power and pumped storage is put forward. In this mode, wind power is the main source while gas-fired power and pumped storage act as an adjustment to smooth wind power output. The model takes the maximum benefit of joint operation as the objective and adopts mixed integer linear programming. Through an example, the economic benefit of the combined operation mode of wind-gas-pumped storage is calculated. The results show that this operation mode not only improves the utilization of wind power, but also improves the efficiency of power grid operation.