Controller Optimization

Reflex-based controller for assistive devices using parameter optimization

Controller Optimization Demo

Controller optimization in MyoAssist enables optimization of a reflex-based musculoskeletal model controller combined with exoskeleton controllers. Using CMA-ES (Covariance Matrix Adaptation Evolution Strategy), this framework can produce controllers that achieve diverse performance objectives.

Optimization Workflow

  1. Setup: Configure your musculoskeletal model and exoskeleton controller
  2. Define Objectives: Specify environment configuration, cost functions, and optimization criteria
  3. Optimize: Run CMA-ES optimization to find optimal controller parameters
  4. Monitor Progress: Track CMA-ES progress and output cost values
  5. Analyze Results: Evaluate results and visualize performance

Key Features

  • Reflex Control Optimization: Optimize reflex-based controllers using CMA-ES
  • Exoskeleton Control Testing: Design, deploy, and optimize controllers for various assistive devices
  • Result Analysis: Built-in tools for processing and visualizing optimization results

Key Scripts

  • run_ctrl_minimal.py: Simple reflex control testing with random parameters
  • run_ctrl.py: Full simulation with video generation and parameter loading
  • run_optim.py: CMA-ES optimization runner for controller tuning
  • run_eval.py: Results evaluation and analysis

Getting Started

Learn the basics of reflex control and start your first optimization

Additional Topics and Tools

Customize cost functions and analyze optimization results

Getting Started

Codebase Structure 

ctrl_optim/
├── run_ctrl_minimal.py          # Quick testing script
├── run_ctrl.py                  # Main simulation script
├── run_optim.py                 # Optimization runner
├── run_eval.py                  # Evaluation script
├── results/
│   ├── evaluation_outputs/      # Simulation videos and outputs
│   ├── optim_results/           # Optimization results
│   └── preoptimized/            # Pre-optimized controllers
├── ctrl/                        # Controller implementations
│   ├── reflex/                  # Reflex controller modules
│   └── exo/                     # Exoskeleton controller modules
└── optim/                       # Optimization framework
    ├── cost_functions/          # Cost function implementations
    ├── config/                  # Configuration files
    └── training_configs/        # Training configurations

Basic Reflex Control

Start with the minimal script to run reflex control:

cd ctrl_optim
python run_ctrl_minimal.py

This script:

  • Creates random control parameters (77 parameters for 2D reflex controller)
  • Runs a 5-second simulation with default settings
  • Reports walking duration

Table of contents