Font Size: a A A

Research On Robot Binocular Vision-Based Learning By Demonstration And Obstacle Avoidance Trajectory Planning

Posted on:2020-05-16Degree:MasterType:Thesis
Country:ChinaCandidate:X Y LvFull Text:PDF
GTID:2428330572970220Subject:Mechanical and electrical engineering
Abstract/Summary:PDF Full Text Request
Traditionally,robots perform desired tasks with the aid of end-users analytically decomposing and manually programming.Robots can only follow program instructions to move,which bring great difficulty to non-professional users.However,with the development of robotics,human-computer interaction has become the future development trend.Therefore,it is significance to study a robot learning method that allows people without programming ability to manipulate the robot freely.This paper proposes a robot vision-based learning by demonstration(LbD)method.Combining visual sensing technology and robot learning by demonstration technology,this method allows a robot to learn to how to complete a new task.Under the framework of robot vision-based LbD,the robot can automatically "observe" the demonstrator's demonstrations for recording the pose information of the object,and then generate the demonstrated trajectories according to this pose information.And robot also can avoid obstacles autonomously that do not appear in the original demonstrations,but it would interfere with the original demonstrated trajectory.Specific research content includes:(1)I design and build a binocular vision detection platform,and use the parallax principle of the binocular vision camera to find the three-dimensional space coordinates of the object center.Binocular vision is composed of two monocular cameras.I perform binocular stereo vision correction to adjust the left and right camera poses to the best shooting position.Taking the transfer robot as an example,during the process of demonstrating,this platform can capture images,and a series of image processing methods are adopted to process these images such as Gaussian filtering,edge detection,open operation,ROI region of interest selection.These image processing methods can assist robot to extract the three-dimensional coordinates of the center of the box which using for generating the demonstrating trajectory.(2)I complete the three dimensional curve fitting interpolation with dimension transformation for generating demonstration trajectory.The three dimensional curve fitting interpolation is usually used to generating a 3D surface.So I convert 3D discrete space points into 2D plane points for interpolation and reconstruct the 3D demonstration curve.(3)Under the surveillance of the binocular vision camera,if there is an obstacle blocking the demonstration trajectory,under the framework of robot vision-based LbD,robot could plan the obstacle avoidance trajectory autonomously.For the captured object,the robot uses the canny operator for edge extraction.When robot identifies the object is an obstacle,robot would independently plans the obstacle avoidance trajectory to avoid the non-safe domain.(4)Experimental verification.A demonstrator carries a box to demonstrate the move trajectory,and the binocular vision camera captures the picture of the box.Then the control program generates the demonstrated trajectory and independently plans the obstacle avoidance trajectory.Finally robot would reproduce the new trajectory and the error analysis is carried out on the new trajectory.The experimental results show that the proposed method can make the robot complete some simple task by vision-based LbD and have simple obstacle avoidance function.So this paper provides an effective method for the research of robot vision-based learning by demonstration.
Keywords/Search Tags:Robot, Vision-based LbD, Binocular vision, Depth information extraction, Trajectory planning
PDF Full Text Request
Related items