robotology / icub-tech-support

Virtual repository that provides support requests for individual robots
GNU General Public License v2.0
20 stars 2 forks source link

ergoCub 1.1 S/N:001 – Angle offset in the forward kinematic chain reconstruction, from head to soles #1873

Open SimoneMic opened 1 month ago

SimoneMic commented 1 month ago

Robot Name 🤖

ergoCub 1.1 S/N:001

Request/Failure description

I've noticed that there is an angle offset in some joint of the kinematic chain: starting from the head, to the robot feet soles. This can be seen immediately by portraying the camera 3D pointcloud and the robot TFs, with the robot standing still on the floor. In this situation, the camera point on the floor are not aligned with the feet frames.

I thought about some possible causes:

This doesn't happen in simulation

Detailed context

I have used ROS2 and the package robot_state_publisher to reconstruct the transforms, while using the urdf from ergoCubSN001. Here are the result: Screenshot from 2024-07-16 18-53-48 Screenshot from 2024-07-16 18-53-57

The purple floor that you can see is the map, and it's lying on the feet soles plane. As you can see, the pointcloud reconstructed from the camera depth is penetrating the ground floor.

This, however, doesn't happen in simulation: Screenshot from 2024-08-06 18-57-36

Screenshot from 2024-08-06 18-57-52

I am not expecting the pointcloud to be perfectly aligned with the floor on the feet, but this error is to big

Additional context

No response

How does it affect you?

This impacts all the measurements taken from the head and transformed in any other reference frame from below the neck. Giving thus wrong quantities and introducing an error.

SimoneMic commented 1 week ago

Pointcloud_alignement_SN002.webm

I tested on ergoCubSN002 the same setup: as you can see from the video above the pointcloud is well aligned. The ground floor lies on the plane of the feet soles.

So this means that it's an hardware issue limited to ergoCubSN001