Using Bullet for tendon-driven robot simulation

[wittmeis]

Hi,

I am working on the EU-funded robotic research project "ECCEROBOT" which goal it is to build the first anthropomimetic robot (see http://www.eccerobot.eu).

The robot is actuated by tendon-driven actuators - namely a dc motor in series with a kiteline and a spring. By winding up the kiteline the artificial muscle gets "flexed" (http://eccerobot.org/home/robot/actuatorsubsystem). As the physical robot is extremely hard to track, we implemented a simulation environment for it based on Bullet and Coin3D (see screenshot).

The critical part of the simulation is of course the muscle simulation. So far, I had simulated the muscles "virtually" using Hookes Law and the geometrical anchor distances. This worked quite well but, unfortunately, has some limitations as the direct line between the anchor points of the muscle is no longer the muscle length once the muscle would collide with the structure which results in wrong force vectors and forces.

To compensate for that I started investigating the Bullet SoftBodies for this purpose. That approach looked quite promising in simple examples and I started implementing a "springAnchor" to simulate the springs of the robot actuators for the soft-bodies. However, now in a bigger setup I am experiencing problems with the stability of the simulation, especially with the soft-rigid body collision. The muscles always penetrate the rigid bodies which creates wild fluctuations due to the resulting collision forces I guess (I attached a screenshot). Is there any way around that problem? I've read in other posts that using SoftBody clusters might help. However, when I tried it I could not see any difference. Is it sufficient to call "generateCluster" and add the required flags to the collision detection or is there anything else that needs to be set for it to work properly?

Could it maybe just be a question of setting the right mass ratios as well? I scaled the world by 100 which increased the stability a lot already and played with the masses of the rope but so far without success. The rope weights around 10-50 gramms. The links around 200-300 gramms.

Any help is appreciated!

Thanks a lot,

Steffen Wittmeier

SoftBodyPenetration.png