Hi, I am trying to model a robotic arm (UR5) equipped with an underactuated gripper (robotiq s model). The gripper has 5 motors in total: the main 3 fingers motors are located at the gripperpalm, while the other 2 provides the scissor mode of the 2 finger B and C in attached fig 1. The URDF of the gripper is provided with box links that simulate the hinge actuation bars on the back of the fingers, making possible through a gazebo plugin the underactuated behavior in attached fig 2. Moreover, this gripper can use this method to adapt itself to the object to be be grasped (attached fig 3). As mentioned in the documentation i converted the URDF file of the whole robot+gripper to MJCF using the compile executable. After the first simulations, it is clear that the hinge bars that actuate indirectly the gripper's phalanges are not connected, thus the simulator considers the phalanges joints as free joints (that can be actuated). Is there a proper way of doing this in mujoco? I would be happy even with a fig 2 behavior, ignoring the adaptive feature. Thanks for the help.
There are two ways to do this. The first is to model the linkages explicitly and to introduce equality constraints - which is how the actual system works. But a more efficient way (computationally) is to use fixed "tendons" which are just linear combinations of joint angles. You can put equality or inequality constraints on the tendon lengths, as well as actuate the tendons. Some of the coefficients in the linear combination can be negative, allowing you to model differentials. If you look at the MPL hand model (under Resources, also distributed with MuJoCo HATPIX) it provides an example of under-actuation, although your model is more elaborate. I actually have a very elaborate example of an adaptive under-actuated system called the Luke Hand from DEKA Research, but am not yet allowed to distribute it outside the performer teams for the DARPA HAPTIX program. It may become publicly available at some point though.
Good to hear. The UR5 model may be of interest to many other users actually. So if you could post in on the Forum when you are done fine-tuning it that would be great.
Sure! Though I'm still having some issues in controlling the actuators.. The P controller for joint positions is not perfect, since it is hard for me to find the right kp and damping for every joint. I think I will use the motor tag for torque control, but I'm having a hard time finding the "gear" parameter in the UR5 manufacturer datasheets. I don't like too much the trial and error approach since i would like a pretty plausible model. These should be the motor reduction ratios, right?
@Emo Todorov The model is finally available for the community! http://www.mujoco.org/forum/index.p...bots-ur5-robotiq-s-model-3-finger-gripper.22/