Research on artificial intelligence and robotics for the ecological protection and restoration of waters has increased in importance with the promotion of national green sustainable development strategies. Autonomous Surface Vessel for Water Cleanup (ASV) research has not yet considered the uncertainty of water surface float positioning under complex current fluctuation factors. The result is that ASV has low accuracy in planning water surface float cleaning paths and its efficiency cannot meet the practical application requirements. To solve the above problems, this paper proposes a dynamic path-planning method architecture based on UAV flow velocity prediction for narrow and shallow water environments such as urban rivers and lakes. We have developed a fusion algorithm based on the progressive optimal rapid random search tree (RRT*) algorithm, DWA dynamic window method, and artificial potential field, and we have realized multi-objective water surface floating under uncertain water flow fluctuations. The location prediction of objects and optimum cleaning paths are carried out, and the method verification is carried out through numerical simulations and prototype experiments. Based on simulation and experimental results, the unmanned surface trash cleaning boat is capable of dynamically predicting trash floaters on the water surface under different water flow environments and generating an optimal cleaning path in real-time. It has certain advantages over existing planning methods in terms of accuracy and timeliness and holds significant practical significance.
Pickled mustard tuber (PMT), also known as Brassica juncea var. tumida, is a conical tuberous vegetable with a scaly upper part and a coarse fiber skin covering the lower part. Due to its highly distorted and complex heterogeneous fiber network structure, traditional manual labor is still used for peeling and removing fibers from pickled mustard tuber, as there is currently no effective, fully automated method or equipment available. In this study, we designed an underactuated humanoid pickled mustard tuber peeling robot based on variable configuration constraints that emulate the human “insert-clamp-tear” process via probabilistic statistical design. Based on actual pickled mustard tuber morphological cluster analysis and statistical features, we constructed three different types of pickled mustard tuber peeling tool spectral profiles and analyzed the modular mechanical properties of three different tool configurations to optimize the variable configuration constraint effect and improve the robot’s end effector trajectory. Finally, an ADAMS virtual prototype model of the pickled mustard tuber peeling robot was established, and simulation analysis of the “insert-clamp-tear” process was performed based on the three pickled mustard tuber statistical classification selection. The results showed that the pickled mustard tuber peeling robot had a meat loss rate of no more than 15% for each corresponding category of pickled mustard tuber, a theoretical peeling rate of up to 15 pieces per minute, and an average residual rate of only about 2% for old fibers. Based on reasonable meat loss, the efficiency of peeling was greatly improved, which laid the theoretical foundation for fully automated pickled mustard tuber peeling.