Simulink parameter optimization for walking algorithm
2 views (last 30 days)
Hi, I am currently doing my Master Thesis in Robotics and Bio-mechanics and I've been using Simulink for about 5 months. The goal of my project is to develop a walking algorithm that will be used in humanoid robots. I already created the control model in Simulink but now I have about 20 parameters to optimize in order to achieve a stable, human-like walking pattern. I'm here asking some help of how this can be made, as I am very new to Simulink and never did a work in optimization. My optimization goals is for example, to maximize the number of steps that the humanoid does, or to maximize both the time and distance traveled by the humanoid before he falls. By now my control model, when running, displays a humanoid robot walking and only stops due to constrains violations or if the time of simulation is exceeded.
Thank you in advance :)
Ryan G on 10 Jul 2013
In my opinion, 20 Parameters is a significant optimization problem. I would try to narrow down the parameters required to those that are most effective in changing system dynamics. It also sounds like you are using physical modeling aka simscape aka simmechanics. When you see a constraint violation that means your system (humanoid) is crashing.
If you design your system well, you can avoid this by having him fall over or having the joints hit a limit instead of causing a critical dynamic fault. However, this is not always easy or possible given the time to make the simulation.
That being said, you likely have Simulink Design Optimization at your university. Take a look at some of the videos and examples on that page to get started. Try optimizing for 1 paramater first, then build up to 20. You may want to group the parameters together, if possible, to simplify the optimization.
One last thing, make sure you are near 100% confident in your model before optimizing. If you're model is not built properly, the optimization may never work and you will spend a lot of time optimizing a bad model. How you verify your model is up to you.