Open MarvinSt opened 1 year ago
EricCousineau-TRI
commented
1 year ago Assigning @amcastro-tri for disposition
FWIW: This could be modeled via a holonomic constraint h(q) = y_a(q) - y_d(t, q) = 0. I am not sure if our current coupler constraint API admits something this general.
You can check your workflow by implementing this in a controller, assuming kinematics are far enough away from singularity.
MarvinSt
commented
1 year ago Thanks for the feedback. A custom function is probably the most general implementation, allowing the user to specify the 6 DOFs of this body based on inputs of the connected frame(s). In the case of steering, this would require 2 input frames. Using an auxiliary 1D vertical prismatic joint for the vertical motion, we could do something as follows to express the upright transformation
x,y,z,r,p,y = upright_motion(steering_frame.x, aux_vertical.z)
Within this function one could perform the lookup table interpolation and perhaps a standard component with a lookup table can be provided (up to 2 input dimensions perhaps). Not sure how to deal with the higher order terms though, perhaps taking the numerical derivative from the table.
I saw that Simbody has Mobilizer::FunctionBased and Constraint::PrescribedMotion:
sherm1
Feature request: To model complex kinematics (for example, automotive multi-link suspension geometry, see image below) efficiently for real time simulation, it is quite common to reduce this system to a single vertical and steering DOF, while using a lookup table to describe the motion of the constrained DOFs with respect to these free DOF(s) (i.e. pitch, yaw, roll, x, y depending on vertical movement and steering).
Ideal solution: Custom joint with one or two degrees of freedom (configurable) and a 1d or 2d lookup table, respectively. There is potentially cross coupling between the two inputs which needs to be dealt with (i.e. steering rack induces vertical movement). However, for a first try, I would simply try to implement prescribed motion for pitch, yaw, roll, x, y on vertical motion.
Such a joint could also be used to model other systems like variable ratio and a cam profile.
Alternatives: Have not considered any alternatives, but one could of course cheat using a quasi static transformation after each time step, however, this would not be as accurate of course. Of course, a full MBD solution is also possible, given that the run time cost is acceptable.
Additional context Such a "joint" is I believe possible to implement in libraries like Simbody and Project Chrono, it is also available in certain Dymola libraries designed for modelling automotive suspension kinematics.