WalkingModule+Bimanual Manipulation Integration in Simulation and Real Robot.
Closed this issue · 7 comments
This issue is to keep track of the integration of the Walking module with the Bimanual manipulation. With the help of @mebbaid ,we were able to test his module and observe that the robot performs reasonably stable walking with the hands at various poses. Now the next step is to integrate the walking module with @Woolfrey 's module.
Current issue I face is that when i called the grasping module after doing prepareRobot and startWalking, the arms flailed and yarpmotorgui showed hardware fault on the simulated robot. The progress of the resolution of this issue will be tracked here.
Had a meeting with @mebbaid. He said the walking controller can be modified in a way such that they use references from @Woolfrey's module for the arms and torso like they do when they are using the suit and teleoperation. The modifications will be added to @SimoneMic version/fork of the walking controller (navigation_unicycle branch) most likely through a temporary branch for testing.
Progress as of 5-09: With the help of Mohamed, dirty version of the joint-retargetting has been implemented. There seem to be some segmentation issues + config file issues causing behaviors as seen in the video below, when @Woolfrey's module sends joint references. Fix is underway.
grasp_walk_test4-2023-09-04_14.49.56.mp4
With the help of Mohamed, we added a thrift interface message to bimanual manipulation to send upper bodies joint references to the walking controller. Software integration is complete and the result can be seen in the video below. Sometimes there seem to be joint configs in which the walking controller seems to fail inconsistently which needs be re-tested to re-produce the error reliably, but the software integration is now complete.
bimanual_walk_integration-2023-09-06_17.04.38.mp4
A somewhat successful simulation video combining walking + object manipulation. We move the object relatively left and right and the roll it in two directions.
move_object2-2023-09-07_20.07.35.mp4
With @mebbaid , we run some tests of the integration of bimanual+walking on the real robot. We observed joint tracking errors on the following 4 joints.
L_SHOULDER_PITCH
R_SHOULDER_PITCH
L_SHOULDER_ROLL
R_SHOULDER_ROLL
We think it the low-level PID tuning that is the most likely issue as the errors are somewhat similar in profile. We hope to solve this on Wednesday when Mohamed, @SimoneMic and I have the robot booked for almost the whole day depending on the situation on that day's experiments.
Below is a shot video of a somewhat successful walking+bimanual experiment which was performed yesterday with @mebbaid. The robot walks while the instruction is to move the object left is sent to it.
walking_plus_move_left.mp4
Software wise, integration is complete in both simulation and real robot.. Now the performance needs to be tuned so that we can see better grasping results and hence closing the issue.