force torque sensor reports wild readings at joint limits using ODE

[osrf-migration]

Original report (archived issue) by Shane Loretz (Bitbucket: Shane Loretz, GitHub: sloretz).

---

While writing a tutorial for the force torque sensor on gazebo 7 I noticed that the readings from the force_torque sensor vary wildly when a joint comes to rest at one of its limits.

To reproduce

• gazebo --verbose worlds/force_torque_demo.world
• gz topic --view /gazebo/default/model_1/joint_01/force_torque/wrench
• Apply a force of 500 N on the +Y direction to link_2. The joint will fall over
• Observe that while the joint appears at rest, the readings from the force torque sensor vary wildly.

Expected behavior

• I expect that the forces and torques on the sensor would be constant when the joint is stationary.
• ForceY: +196 N
• TorqueX: -147 Nm

Here are some of the actual readings

• ForceY: 249.9 N, TorqueX: -229.3 Nm
• ForceY: 204.6 N, TorqueX: -160.0 Nm
• ForceY: 233.8 N, TorqueX: -203.9 Nm
• ForceY: 99.1 N, TorqueX: 0.0 Nm

It's almost like the joint is bouncing. The last reading is almost like the sensor was reporting just the force due to gravity on link_1.

[osrf-migration]

Original comment by Shane Loretz (Bitbucket: Shane Loretz, GitHub: sloretz).

---

@scpeters discovered that setting the solver parameter SOR to 1.0 (default 1.3) solves the issue.