Hinge joint can't beat gravity [SOLVED]

[JoeWright]

I'm trying to do a motor hinge joint (based on the custom hinge joint code) but I can't get it to beat gravity.

I've got a collection of bodies that are connected with hinge joints and the code is asking Newton to force the hinge angle to a fixed value (0.0). The problem is, gravity has a bigger effect on the hinges - I just can't get them stiff enough.

Any ideas?

Joe

[JoeWright]

This is my person lying collapsed on the floor. I don't have collision between body parts by design. All hinge joints are trying to contrain to a default zero angle. This means the left leg and arm (green) should be straight but they're actually hanging towards the floor (into the other body parts as I have no collision).

My modification to the hinge code is as follows:

Code: Select all

   desired_angle=0.0;

   if (angle < desired_angle) {
      dFloat relAngle;

      relAngle = angle - desired_angle;

      // tell joint error will minimize the exceeded angle error
      NewtonUserJointAddAngularRow (mpNewtonJoint, relAngle, &global_pin_pivot0.m_front[0]);

      // need high stiffness here
      NewtonUserJointSetRowStiffness (mpNewtonJoint, 1.0f);

      // allow the joint to move back freely
      NewtonUserJointSetRowMaximumFriction (mpNewtonJoint, 0.0f);


   } else if (angle > desired_angle) {
      dFloat relAngle;

      relAngle = angle - desired_angle;
      
      // tell joint error will minimize the exceeded angle error
      NewtonUserJointAddAngularRow (mpNewtonJoint, relAngle, &global_pin_pivot0.m_front[0]);

      // need high stiffness here
      NewtonUserJointSetRowStiffness (mpNewtonJoint, 1.0f);

      // allow the joint to move back freely
      NewtonUserJointSetRowMinimumFriction (mpNewtonJoint, 0.0f);

   }


Joe

[JoeWright]

I've solved this by switing to linear rows instead of the angular row:

Code: Select all

   q0=global_pin_pivot0[1].Scale(cos(desired_angle))+global_pin_pivot0[2].Scale(sin(desired_angle));
   q0=q0.Scale(MIN_JOINT_PIN_LENGTH);
   q0+=global_pin_pivot0.m_posit ;
   q1=global_pin_pivot1.m_posit + global_pin_pivot1[1].Scale(MIN_JOINT_PIN_LENGTH);

    NewtonUserJointAddLinearRow (mpNewtonJoint, &q0[0], &q1[0], &global_pin_pivot0.m_up[0]);
   NewtonUserJointAddLinearRow (mpNewtonJoint, &q0[0], &q1[0], &global_pin_pivot0.m_right[0]);

Julio, I think you've mentioned in the past that angular rows are not as powerful as linear rows. That's what I'm finding here. Why is that?

Cheers, Joe

[JoeWright]

I've found that the above code doesn't always work. I've cured it by changing the constraint to be along the tangent to the child moving axis, in the moving plane. This is the cross of the child moving axis and the hinge axis:

Code: Select all

   q0=global_pin_pivot0[1].Scale(cos(desired_angle))+global_pin_pivot0[2].Scale(sin(desired_angle));
   q0=q0.Scale(MIN_JOINT_PIN_LENGTH);
   q0+=global_pin_pivot0.m_posit ;
   q1=global_pin_pivot1.m_posit + global_pin_pivot1[1].Scale(MIN_JOINT_PIN_LENGTH);

   dVector tangent(q1 * global_pin_pivot0[0]);
    NewtonUserJointAddLinearRow (mpNewtonJoint, &q0[0], &q1[0], &tangent[0]);

Its also useful to avoid stuck conditions at the limits. For example, I use this code that makes sure the desired angle is only clamped if the acceleration is trying to make it more extreme:

Code: Select all

   desired_angle=mLastHingeAngle+mHingeAcceleration*kTimeStep*TWO_PI;
   if (desired_angle<min_angle && mHingeAcceleration<0.0)
      desired_angle=min_angle;
   if (desired_angle>max_angle && mHingeAcceleration>0.0)
      desired_angle=max_angle;