Hello,

I am new to Houdini and I like an idea to make animation completely in DOP using bullet physics engine (not using classical bones rigging).
Currently I am animating a robot using 4 “Cone Twist Constraints” with motors. Two motors in robot's foot control robot twist and tilt. Two other motors are in shin and hip.

For example, I can make the robot jump by stretching rapidly it's leg. How far it can jump is controlled by tilt of the robot and
strength of the motors in robots shin an hip. I also use some anchors to hold the robot vertically when it is necessary.

I am happy how simulation looks, but it is very hard to control it. Basically I need to find correct settings for each motor to make the robot jump and land where I need it to land. And it is much harder if robot needs to make 2-3 jumps in a row.

My final goal is to make short animation film where robot can jump on some objects, perhaps, play volleyball, e.t.c.
And I am not sure if it is a good idea to make this film in this way. It involves a lot of experimenting with constraint settings and each time simulation begins from start.
Is it valid way to animate a robot? Are there any suggestions how could I improve this process?






Thanks.

Very cool!
This might be a good opportunity to wedge a whole bunch of different constraint settings and iteratively refine until you find best values. Over several generations you might converge at some best general purpose scheme, or perhaps an adaptive one? It could be interesting to leverage PDG for this kind of research.
Is your way a valid approach? It depends really on what your personal goal is. If you just want a nice little animation of a single robot in a short time, then no. But, to learn some fun and difficult technical tasks and get a fun short film from your effort, then for sure, why not?
I think this kind of approach really lends itself to tasks that would be prohibitive to animate. So, if you manage to get a single robot to predictably play volleyball, then you should be ambitious and aim for 10 robots playing volleyball together.

Have you looked into machine learning/genetic algorthms? Sort of like what Ciaran was saying, but guided by fitness evaluation or neural networks.

https://karlsims.com/evolved-virtual-creatures.html [karlsims.com]

I thought about it, but haven't looked yet. I suppose there are some ML libraries available for Python. Can any of those be used in Houdini?