My simulation explodes or behaves weirdly

[omichel]

How can I fix it?

[omichel]

The explosion is usually caused by inappropriate values passed to the physics engine (ODE). There are many things you can be try to improve the stability of the simulation (adapted from ODE's User Guide):

1. Reduce the value of WorldInfo.basicTimeStep. This will also make the simulation slower, so a tradeoff has to be found. Note that the value of the control step (wb_robot_step(TIME_STEP)) may have to be adapted to avoid warnings.
2. Reduce the value of the JointParameters.springConstant and JointParameters.dampingConstant fields or avoid using springs and dampers at all.
3. Avoid large mass ratios. A Joint that connects a large and a small mass (Physics.mass) together will have a hard time to keep its error low. For example, using a Joint to connect a hand and a hair may be unstable if the hand/hair mass ratio is large.
4. Increase the value of WorldInfo.CFM. This will make the system numerically more robust and less susceptible to stability problems. This will also make the system look more spongy so a tradeoff has to be found.
5. Avoid making robots (or other objects) move faster than reasonably for the time step (WorldInfo.basicTimeStep). Since contact forces are computed and applied only at every time step, too fast moving bodies can penetrate each other in unrealistic ways.
6. Avoid building mechanical loops by using Connector nodes. The mechanical loops may cause constraints to fight each other and generate strange forces in the system that can swamp the normal forces. For example, an affected body might fly around as though it has life on its own, with complete disregard for gravity.
7. Avoid thin objects. The problem with thin objects is the following: since the simulation time increases step-by-step, if the time step is too large and a thin object is moving fast, it is possible that at step t, the thin object does not collide and at step t+1 it is already completely inside another object. In that case, ODE (the physics engine of Webots) might not know on which side to push back the thin object and this might result in big instabilities. Similarly, you can think of a bullet fired into a wall. If the simulation step is too large, it is completely possible that the bullet goes through the wall without even noticing it. At time t, it will be on one side of the wall and at time t+1, it will be already on the other side, so no step is producing a collision.