The main purpose of this page is to describe the working principle behind the simulation runner. This entry point enables the execution of arbitrary motion profiles on all actuators and measures the behavior of the quadrupole. This script uses the inverse kinematic function obtained by training a feedforward neural network on a big dataset of joint positions and associated quadrupole positions/orientations.
Location
The models are located in project_root/Simulation scripts
- Sim_runner_full_nn.m: Runs the 5dof model with the inverse kinematics computed through the inference of a 40 hidden layers FF NN
Script's description
Overview
This script interfaces with the SCU model, loads pre-trained neural networks for inverse and direct kinematics, generates and simulates actuator position signals, and performs error analysis on the simulated data. This can be modified to obtain any arbitrary motion.
Key Steps
Initialization:
- Clears variables, determines the script's path, and constructs paths for the SCU model and neural network files.
Load Neural Networks:
- Loads inverse and direct kinematics neural networks from specified .mat files.
Generate and Configure Signals:
- Sets simulation parameters (10 seconds, 1000 points), creates time vectors, and initializes actuator positions.
- Defines trajectory parameters and generates positions and orientations over time.
Calculate Joint Positions:
- Uses the inverse kinematics neural network to compute joint positions for the desired quadrupole positions/orientations.
- Plots the command positions and orientations, as well as the joint positions.
Run Simulation:
- Loads and simulates the SCU model using the computed joint positions.
- Captures quadrupole position and orientation data.
Analysis:
- Plots quadrupole positions and orientations, performs error analysis, and visualizes cross-correlations between positions and orientations.
Results
Motion on X, Y, PITCH, ROLL
Motion on X and Y with trapezoidal motion profile
