First Plan Then Evaluate: Multi-Target Planning with Post-Planning Success Evaluation Improves Learning-Based Grasping Pipeline

Key takeaways

1) Plan trajectories to all the grasp targets using a motion planner that supports efficient planning to multiple different targets (e.g. cuRobo)

2) Don't reject any of the planned trajectories based on the distance to the target (IK error)

    3) Evaluate grasps from the planned trajectories for grasp success, not the targets for the motion planner