Challenge 2: Grasp needle and drive through tissue
Move large needle driver to grasp the needle and then move the needle tip to the target and drive the needle through the tissue until the tip exits. The accuracy of the simulated robot will be comparable to that of a real robot and thus visual feedback would be required to ensure accurate performance.
The needle should be grasped near its base (i.e., back third of needle) with the gripper orthogonal to the plane of the needle. However, the evaluation metrics do not consider the needle grasp pose, so any pose that produces successful results is acceptable.
- Initial ground-truth pose of needle
- Ground-truth poses of first target entry and exit on phantom
Note that the entry and exit points are specified as poses (transforms). See this figure. These coordinate frames will always be defined so that the XZ planes are aligned. The expectation is that the needle trajectory will primarily be a rotation about the Y axis (i.e., in the XZ plane), though the evaluation metric does not require this.
Each entry will be tested with multiple trials, where the needle position and target positions will vary for each trial. In addition, some kinematic parameters of the robot will vary from the nominal values to emulate robot inaccuracy. Specifically, the simulator will use the nominal kinematic parameters for inverse kinematics (e.g., to convert Cartesian goals into joint goals), but then will use the perturbed kinematic parameters in the forward kinematics to compute the Cartesian trajectory of the simulated robot. The kinematic parameters that can vary are the offsets for the first two joint angles (up to ±5 degrees) and the third joint translation (up to ±5 mm). For some trials, the kinematic parameters will have little or no error, thereby allowing the task to be completed without visual feedback, whereas for other trials, visual feedback will be necessary to successfully complete the task. Stereo endoscope video will be provided at 1080p resolution at 30 fps.
Entries will be judged based on the time the algorithm required to perform the task and the achieved accuracy. Accuracy will be measured by the distance between the needle trajectory and the target entry and exit positions, and the amount of needle that is visible beyond the exit point at the end of the task. User scripts should publish the specified ROS topic or call the specified Python method to indicate task completion. All entries must be within ±2.5 mm of the target entry/exit and at least 5 mm of the needle tip must be visible. Note that the squares surrounding the entry/exit points have dimensions 5 mm x 5 mm (0.05 x 0.05 simulation units, see figure below), so the needle must pass within these squares. All entries that pass the accuracy threshold will be ranked based on completion time. Time will be measured from when the user script is started until the task completion message is received.
The evaluation script for this challenge is in the GitHub repository and can be run as follows (use
python3 as appropriate):
python evaluation.py -t <team_name> -e 2
See also the GitHub Discussions forum.