Running Reflex Control

This document explains how to run the reflex controller using the MyoAssist framework. There are two main scripts available:

  1. run_ctrl_minimal.py: A simple script for running a reflex control simulation with random parameters
  2. run_ctrl.py: A more in-depth script for reflex control simulation with video generation

Running run_ctrl_minimal.py

Usage 

cd ctrl_optim
python run_ctrl_minimal.py

What it does

  • Creates 77 random control parameters. This is the 2D reflex total. See Reflex Control for the counts of the other modes.
  • Runs a 5-second simulation with default settings
  • Reports walking duration
  • No video generation or file outputs

Walking duration will vary each time run_ctrl_minimal.py is rerun since the parameter vector is randomized.

Configuration

The script uses these default settings:

  • Simulation time: 5 seconds
  • MSK model: myolegs22 (22-muscle 2D lower limb)
  • Device: Tutorial_L1
  • Initial pose: walk_left
  • Slope: 0 degrees (flat ground)
  • Exoskeleton: Disabled
  • Control parameters: Random normal distribution

The environment is composed from msk_key / device_key and an optional terrain. See Defining an Environment.

Output

The script prints the walking duration to the console:

Walking duration: 0.350 seconds

Running run_ctrl.py

Usage 

cd ctrl_optim
python run_ctrl.py

Method 1: Loading from Optimization Results

To visualize a controller that you have already optimized:

# In run_ctrl.py, set:
LOAD_FROM_FILE = True
PARAMS_FILE_PATH = "results/optim_results/results_folder/parameters.txt"

The script will automatically:

  • Load the optimized parameters from the .txt file
  • Find the corresponding configuration file in the same directory
  • Reconstruct the environment used during optimization

The environment is rebuilt from the config saved next to the results (the .bat/.sh inside the results folder), not from training_configs/.

The saved config must use the current registry-key format (--msk / --device). Result folders produced before the assist_sim registry-key rework carry the old --musc_model / --model flags and will fail with:

myoLeg_reflex requires explicit msk_key and device_key … got msk_key=None, device_key=None

To re-run an old result, regenerate it, or update its saved config to --msk / --device. The maintained configs live in ctrl_optim/optim/training_configs/ (the bundled tutorial_example snapshot has been regenerated to match).

By default the script ships with LOAD_FROM_FILE = True, so cd ctrl_optim then python run_ctrl.py (the Usage command above) runs this path as-is.

Method 2: Manual Configuration

To create a new simulation from scratch:

# In run_ctrl.py, set:
LOAD_FROM_FILE = False

# Manual settings:
SIMULATION_TIME = 5       # seconds
SLOPE_DEG = 0             # env slope in degrees
MSK_KEY = "myolegs22"     # human MSK registry key (see `python -m assist_sim list`)
DEVICE_KEY = "Tutorial_L1"  # assist_sim device key
EXO_BOOL = True           # Enable/disable exoskeleton
USE_4PARAM_SPLINE = True  # Use 4-parameter spline for exoskeleton
N_POINTS = 4              # Number of points for n-point spline
MAX_TORQUE = 100          # Maximum exoskeleton torque

Run python -m assist_sim list for the valid MSK / device keys and their compatibility. See Defining an Environment for the full env-spec.

Environment Initialization

This part of the script creates the environment from your configuration:

if LOAD_FROM_FILE:
    # Load from optimization results
    env, config, _ = load_params_and_create_testenv(
        results_dir=results_dir,
        filename=filename,
        bat_file_path=bat_file_path,
        sim_time=SIMULATION_TIME
    )
    print_config_summary(config, title="Loaded Configuration")
    
else:
    # Use manual settings
    config = {
        'mode': '2D', 'init_pose': 'walk_left', 'delayed': False,
        'slope_deg': SLOPE_DEG, 'msk_key': MSK_KEY, 'device_key': DEVICE_KEY,
        'exo_bool': EXO_BOOL, 'use_4param_spline': USE_4PARAM_SPLINE,
        'n_points': N_POINTS, 'max_torque': MAX_TORQUE
    }
    
    # Calculate control parameters
    if config['exo_bool']:
        spline_params = 4 if config['use_4param_spline'] else (config['n_points'] * 2)
    else:
        spline_params = 0
    control_params = np.ones(77 + spline_params)
    
    env = myoLeg_reflex(sim_time=SIMULATION_TIME, control_params=control_params, **config)
    print_config_summary(config, title="Manual Configuration")

This manual example is 2D. It sets mode to 2D and builds 77 reflex parameters. To visualize a 3D, bilateral, or amputee controller, use Method 1 and load it from its optimization results. Method 1 reads the saved config and rebuilds the exact environment and parameter layout. See Reflex Control for the parameter count of each mode.

Simulation and Video Generation

The script runs the simulation with frame recording for video creation:

  1. Camera Setup: Configures a free camera that follows the model’s movement
  2. High-Resolution Rendering: 1920x1080 resolution at 100 FPS
  3. Progress Tracking: Shows progress bar during simulation
  4. Frame Collection: Captures each frame for video compilation

Example Output 

Outputs will be saved to: results/evaluation_outputs/run_ctrl_date_time
Running 500 timesteps...
Progress: |████████████████████████████████████████████████████| 100.0% (500/500)
Video saved: simulation_regular.mp4 (1920x1080)
Video opened in new window: simulation_regular.mp4
All outputs saved to: results/evaluation_outputs/run_ctrl_date_time
Simulation completed successfully!

Troubleshooting

Common Issues

  1. Video not generating: Video export uses imageio with an ffmpeg backend. Ensure it is installed: 
    pip install imageio imageio-ffmpeg

  2. Model / device not found: Environments are composed from registry keys, not bundled models/ files. Run python -m assist_sim list to see valid msk_key / device_key values (see Defining an Environment).

  3. Parameter file not found: Verify the path to your optimization results file

  4. Camera issues: The script automatically handles camera positioning, but you can modify the camera behavior in the script if needed

Performance Notes

  • For faster rendering, reduce the video resolution in the script or adjust SIMULATION_TIME
  • The script automatically terminates early if the model falls or fails