Throwing-trajectories-dataset

Dataset used in the article 'Robot Skill Learning in Latent Space of a Deep Autoencoder Neural Network'

Dataset

Throwing_trajectories.mat contains data structure:

robot_throws

• .robot_data (kinematical data for simulation - see Appendix D)
  • .L are lenghts of links,
  • .base is robot base position,
  • .dt is time step of generated trajectories,
• .qPath{:} (array of 9824 throwing trajectories, each organized in a matrix with 10 columns)
  • [:,1] are times ,
  • [:,2:4] are joint positions ,
  • [:,5:7] are joint velocities ,
  • [:,8:10] are joint accelerations ,
• .targets (matrix of throwing targets)
  • targets[:,1] are target positions ,
  • targets[:,2] are target positions ,
  • targets[:,3] are hit angles ,
• .DMP{:} (array of 9824 dynamic movement primitives (DMPs) describing the throwing trajectories)
  • .N is number of radial basis functions,
  • .dt is time step of DMP integration,
  • .a_z is constant ,
  • .a_x is constant ,
  • .c are center positions of radial basis functions,
  • .sigma_2 are squared variances of radial basis functions ,
  • .w is weight matrix ,
  • .tau is time constant ,
  • .goal is ,
  • .y0 is starting point .

Matlab support code for dataset

kinematicsJacobian.m - calculates the end-effector position and the Jacobian matrix from the robot data in 'robot_throws.robot_data' and the selected joint position .