Pick-Place-Plug Manipulations Under Uncertainties Based On Deviation-fusion Mechanism

Robots are automation equipments that integrate advanced technologies in multiple disciplines such as mechanics,electronics,control,sensing and artificial intelligence.Affected by uncertainties,existing robot applications have high requirements for work scalability and the structural characteristics of the environment,which severely restrict the development of the robotics industry.How to eliminate the influence of uncertainties has become a research problem worthy of investigation.The existing approaches to adapt to uncertainty tend to minimize uncertainties or establish feedback mechanisms to make adjustments when the impact of uncertainty occurs.However,this leads to three problems:an over-reliance on accuracy;lagging feedback mechanisms;and a gap between algorithms and reality.In response to the aforementioned challenges,the thesis focuses on two key issues:finding and designing suitable deviation-fusion mechanisms,and constructing relationships beween algorithm parameters and uncertainty in the system.By starting from contact-rich scenarios such as peg-in-hole and Robotic 3d BinPacking,the thesis explores cost-effective solutions by introducing feedback mechanisms based on interactions between target objects and execution,objects and the environment,perception and execution,all within the existing framework of object manipulation.By replacing the feedback mechanism of traditional planning algorithms,which relies on high-precision sensors and high-frequency control devices,with the feedback mechanism of force deformation of soft components,the robot is able to improvise when contacting obstacles,while the perception deviations and execution deviations cannot be ignored.Based on the deformation-feedback mechanism,a new motion planning problem,the deviation-fusion path planning problem,is proposed,and a set of relevant algorithms for the deviation-fusion process is developed.The mechanism underlying the significant advantage of the relevant algorithm for the deviation-fusion planning problem in the peg-in-hole scenario with unpredictable errors are revealed.Results show that the new algorithm can be widely applied to a number of scenarios such as block assembly,bin-packing,and peg-in-hole.In response to the lack of solutions for robotic bin-packing of unknown-size and deformable items in the logistic scenarios,the thesis summarizes the difficulties as the existence of unpredictable errors and policy errors.Based on the deformation-feedback mechanism and the force-displacement mechanism,a motion planning algorithm for unpredictable errors in contact-rich scenario and an algorithm for online bin-packing are proposed.A complete Robotic 3d Bin-Packing system is developed and its performance under the aforementioned uncertainty is verified through both simulation and physical experiments.Finally,in response to the need for adaptability to dynamic environments and error tolerance,the thesis establishes and introduces a local end-to-end feedback mechanism.Through the interaction of the decision-making system,motion planning,and local feedback mechanisms,the robot is able to handle dynamic environments and adapt to unexpected environmental changes from sensor errors,joint clearances,and human interference.