utiasdsl/upright

[RA-L 2023] Fast nonprehensile object transportation with model predictive control on a mobile manipulator.

upright's Issues

Consult how to realize Sim.

• Thanks for your open source, your method is so elegant.I want try to recurrent it.
• Follow your step, I have complete the job of install and compile.
• Could you give me a more detailed way to sim this paper.Thank you very much!

Help Needed with Smooth Trajectory Following in Robotic Arm Control

Dear adamheins,

I am using your InertialAlignment cost function to achieve balance of the end-effector object. In Rviz, it appears to work well, but I encounter some issues when implementing it on a real robotic arm. How do you achieve such smooth trajectory following?

I am facing challenges with the following:

1. Handling Acceleration in MPC State:

   • In the MPC state, I have position, velocity, and acceleration, with jerk as the input.
   • To follow the trajectory, how should I handle the acceleration? Using delta_velocity / dt seems unstable, possibly due to the instability of ros::Time.

2. Trajectory Following Method:

   • What method do you use for trajectory following?
   • I am using Pinocchio for gravity compensation and applying Kd and Kp to follow position and velocity.
   • However, the performance is poor, likely due to inaccuracies in the URDF model and the complexities introduced by friction and other real-world factors.

Could you please share your approach for achieving such smooth trajectory following in real robotic arms?

I appreciate your time and look forward to your response. Thank you very much!

Best regards,

for static obstacle avoiding

I read the code and have a question : you realize the static obstacle avoiding by add the obstacle urdf to the robot urdf? if yes , how do you link these two urdf , which one is the base_link and how to create the collison_pair?

About MPC Solver Method ： SLQ and SQP

Hello, sorry to bother you again. Do you know what practical problems the OCS2 SQP (multiple shooting) and SLQ (DDP) are suitable for? I see that only the quadruped robot example uses the SQP option, while the other examples use the SLQ algorithm of DDP. Also, the useFeedbackPolicy option is set to false for SLQ and true for SQP.

I'm not sure why this is the case, but I noticed that you are using the SQP algorithm. What was your reason for choosing this? Additionally, you changed SensitivityIntegratorType::RK2 to SensitivityIntegratorType::RK4. Why is that?

simulation can not run

i can not run the simulation, could you please give a docker about this project, or give me the whole install process, it confuse me a week!!! Thanks

Ask about .py and .cpp 's function.

Hello!
I see both cpp and python code in your upright_control and upright_ros_interface,
Is the python code only used to simulatewith Pybullet, and can the python code be removed if other simulation tools are used, such as Webots and Gazebo?
Hope to hear from you!!

Question About the Real

Regarding the real machine experiments: I want to deploy your code on my experimental equipment, so I read your 'mrt_node.cpp'. I have a few questions:
Why is the 'Kalman filter' applied to the mobile manipulator and not to the projectile?
What do 'mpc_plan_pub' and 'cmd_pub' publish? I couldn't find any corresponding subscribers.
Since cmd_pub already exists, what is the code from 'this line' doing? Why define cmd_vel again?
And why use evaluatePolicy()? difference from rolloutPolicy()?