Jerky joints

Gaming and Visualization Technologies APIs Physics Modeling (closed)
1

I’m working on a robot simulation using the PhysX Candy Wrapper (physx system software v 9.11.0621) and am having some serious trouble controlling the joints (RevolutionJoint) with motors.

First, I know that the Candy Wrapper may not be widely used here, and it may even be the source of the problem, but I’m just checking here to see if there are any common mistakes I may be making.

It seems that no matter what kind of signal I send to the joints, they kind of flap about wildly. They stay within the limits I have specified, but their velocity seems to be very unrelated to the velTarget I’m giving to the motor for each joint. I have tried a wide variety of max joint speeds, forces, etc. and can’t seem to figure this out. I have also tried setting the spring and damping constants (in an attempt to add a kind of friction to the joints), and these seem to have absolutely no effect whatsoever. Below is some code samples of how I’m creating and attempting to manipulate the joints. It is a small subset of the overall code. Keep in mind that while this is in C#, much of the task of creating bodies and joints is roughly the same.

Any guidance or tips are greatly appreciated. Thanks in advance.

BodyDesc shoulder1BD = new BodyDesc();
shoulder1BD.SolverIterationCount = SOLVERITERATIONCOUNT;
BoxShapeDesc shoulder1BSD = new BoxShapeDesc(new Vector3(4 * SCALE, 4 * SCALE, 2 * SCALE));
shoulder1BSD.MaterialIndex = bodyMaterial.Index;
shoulder1BSD.SkinWidth = SKINWIDTH;
ActorDesc shoulder1AD = new ActorDesc(shoulder1BSD);
shoulder1AD.Body = shoulder1BD;
shoulder1AD.Density = DENSITY;
shoulder1AD.GlobalPosition = new Vector3(body.GlobalPosition.X - 7, body.GlobalPosition.Y, body.GlobalPosition.Z);
Actor shoulder1 = scene.CreateActor(shoulder1AD);
shoulder1.Name = SHOULDER1NAME;

        BodyDesc upperLeg1BD = new BodyDesc();
        upperLeg1BD.SolverIterationCount = SOLVERITERATIONCOUNT;
        BoxShapeDesc upperLeg1BSD = new BoxShapeDesc(new Vector3(10 * SCALE, 2 * SCALE, 2 * SCALE));
        upperLeg1BSD.MaterialIndex = bodyMaterial.Index;
        upperLeg1BSD.SkinWidth = SKINWIDTH;
        ActorDesc upperLeg1AD = new ActorDesc(upperLeg1BSD);  
        upperLeg1AD.Body = upperLeg1BD;
        upperLeg1AD.Density = DENSITY;
        upperLeg1AD.GlobalPosition = new Vector3(body.GlobalPosition.X - 14, body.GlobalPosition.Y, body.GlobalPosition.Z);
        Actor upperLeg1 = scene.CreateActor(upperLeg1AD);
        upperLeg1.Name = UPPER1NAME;

RevoluteJointDesc jd1 = new RevoluteJointDesc();
jd1.SetActors(body, shoulder1);
//jd1.MaxForce = MAXJOINTFORCE; // if we speciy no max force, the joint becomes unbreakable, which is what we want
jd1.GlobalAxis = new Vector3(0, 1, 0);
jd1.GlobalAnchor = new Vector3(body.GlobalPosition.X - 5 * SCALE, body.GlobalPosition.Y, body.GlobalPosition.Z);
jd1.Motor.CopyFrom(motor);
RevoluteJoint rj1 = (RevoluteJoint)(scene.CreateJoint(jd1));
rj1.Limits.High.Value = (float)(Math.PI / 4);
rj1.Limits.Low.Value = (float)(-Math.PI / 4);
rj1.Flags = RevoluteJointFlags.MotorEnabled | RevoluteJointFlags.LimitEnabled | RevoluteJointFlags.SpringEnabled;
rj1.Spring.Damper = DAMPINGCOEFFICIENT;
rj1.Spring.Spring = SPRINGCOEFFICIENT;

((RevoluteJoint)joints[i + 4]).Motor.VelTarget = (outputs[i] * 2 * MOTORSPEED) - MOTORSPEED;

2

Hi,

I’m not familiar with Candy Wrapper, but it does sound like this might be a problem in the wrapper code. I’ve used the joint motors successfully in the past. You should also check the moment of inertia of the actors connected by the joint.

Perhaps you could connect this application to PhysX Visual Debugger, it might reveal something useful.

–Mike

3

@physneuro

How are you coming along with your program?
I’m also working on a robotics simulator. My joints were shaking in the beginning too, and I had to play with everything that was adjustable. Skin depth, density, kinematic vs dynamic, friction, etc. I never achieved a 100% satisfactory result but I fiddled with it long enough to get it pretty quiet with a special combination of all these aforementioned things. It seemed to depend on the particular model with its own parameters and each run with its applied forces. How is yours coming along?

Shanyi.

4

Hi,

If the joints do not behave as expected (what is there behaviour then?), then why do you use some formula for the velocity. Use at first just a simple constant value to determine the result.

I do not know a lot about the predefined joint types, we only work with D6Joints, which need to be configured yourself. But looking in the PhysX 3.x guide I can see that you can only set the velocity and force limit. So what is the “stiffness” of the joint? In long changes of links I can imagine that such drives have a bad behaviour. The D6 joints have a proportional and derivative drives for which you can define the desired position, stiffness, desired velocity, damping and maximum force. Works very well for our 7dof robots, which can be attached to an actuated boom with another 8dofs for macro movement.

Out of the box thinking:
-can it be that there is a ground plane in which the robot is partially submerged?  Using the PVD helps a lot!
-what is your time step?

I hope this helps