Research On Robot Operation Planning Based On Demonstration Learning And Task Constraints

Due to the serious aging of population and labor shortage,robots are gradually accelerating the pace of entering all walks of life.Robots replace laborers in heavy and boring work,which not only improves the production efficiency of enterprises but also greatly improves the production environment.However,whether the "machine replacement" or "man-machine cooperation" mode is adopted,it requires the robot to be more anthropomorphic and intelligent,and to be close to the natural ability of human beings.The goal of intelligent autonomous robot manipulating system is to build a robot learning system.The robot learns task specification and knowledge from human experience,and abstracts the robot's motion skills,control strategies and task models.Finally,according to the knowledge learned,the robot can autonomously and intelligently complete the operation tasks in different environments.This paper explores a human-machine interactive robot skill learning and autonomous operation framework for multitasking and dynamic environments,and studies the key technologies for task understanding,behavior imitation,behavior optimization,and skill growth for robot autonomous operations,which is one of the urgent problems in the robot field.The small batch and customized production demand promotes the continuous upgrading of robot operation ability,which requires the robot to have the ability of rapid task learning and skill optimization.Traditional programming methods become time-consuming and laborious,and cannot be quickly migrated to new environments and tasks.Intelligent robots must be capable of learning tasks.The primary difficulty of robotic intelligent operation is the fast learning and accurate realization of complex tasks.This paper first studies an efficient robot complex trajectory learning and reproduction framework,which combines unsupervised trajectory segmentation and probabilistic motion primitives to achieve accurate modeling and accurate reproduction of complex trajectories with a small sample size.It can segment complex tasks and process them in segments.The entire learning process of the framework is completely automated and does not rely on prior knowledge or task-specific models.It is proved that more accurate complex trajectory learning can be achieved by our framework,which is of great value in robot polishing,spraying,welding and other practical robot tasks,and greatly expands the application scene of robot imitation learning.Robot intelligent operation not only requires the robot to have strong learning ability,but also requires the robot to have the ability to adapt to the dynamic environment.The robot is required not only to achieve task feature extraction and modeling,but also to avoid moving obstacles to adapt to the dynamic environment while meeting the task feature constraints.An intelligent motion planning framework for safe and task-oriented robots manipulation in dynamic environment is proposed,which combines probabilistic model and real-time safety control.Robot learn task features based on probability models,uses a visual sensor to detect the dynamic environment in real time and work safely in changing scenarios according to task characteristics.This thesis studies the theory of rapid obstacle avoidance of the whole arm.Robot calculates the collision risk with the concept of robot safety domain to realizes the rapid obstacle avoidance response to dynamic obstacles.This thesis studies the opration planning strategy that satisfies the task feature constraints and the safe motion constraints.The robot can change its motion to avoid obstacles under the task constraints to achieve safe operation and task goals.Compared with the traditional CLIK algorithm,the control framework proposed in this paper has a larger obstacle avoidance space and satisfies the robot task constraints.The robot can not only produce the trajectory with humanoid behavior characteristics,but also has strong environmental adaptability and safety.The proposed method solves the problem of robot task learning and generalization in dynamic environment.Security is a prerequisite for intelligent robot operation.However,due to the nature of interaction with the outside world,the security issues of the robot task in contact with the environment is particularly prominent.Therefore,in addition to exploring the problem of motion planning in the intelligent operation of robots,this paper will also focus on the safety of robot force control operation that interacts with the environment.The robot is able to track the desired position and force profile under the task constraints and ensure the safety of operation.In this paper,a multi-priority strategy based on invariant control is constructed to integrate the motion control and force control in the physical contact task into one framework,so as to achieve the tradeoff between the safety requirements and operation performance.The multiconstraint invariant control guarantees the task constraints,achieving interference suppression and operational safety.The expected target of the task is learned by teaching method,and the error exponential convergence regulator is constructed to eliminate the tracking error quickly.Experiments have proved that the framework is competent of different types of physical contact tasks without establishing a complex dynamic model,and can satisfy dynamic environmental constraints to achieve disturbance suppression.The proposed method has better control effect and faster error convergence speed.Finally,the above methods and theories are extended to the important direction of robotic intelligent operations,that is,coordinated operations of human-robot arms.Here,the human-robot dual-arm coordination operation refers not only to the cooperative operation between the human and the robot,but also to the cooperative operation between the two working robot arms.The key to coordination is how to accurately identify the movement intention of the collaborator and make response.As the robot operation is high-dimensional information containing motion and force,the method of data dimensionality reduction is used to embed the high-dimensional demonstration data into the low-dimensional manifold to analyze the highdimensional input information and correlate it with the operation actions.A dual-arm coordination method based on high-dimensional data analysis and collaborative mapping is proposed.The dynamic environment motion planning and safetyconstrained contact operation are integrated into the framework to expand the task boundary of the robot's intelligent operation,which provides technical support for true human-robot integration.Finally,integration testing and verification are conducted through dual-arm collaboration experiments.The experimental results confirm the effectiveness of the proposed robot intelligent operation framework including motion planning,safety control and coordinated operation.