Welcome to muSim’s documentation!

muSim/nuSim is a computational framework for predicting motor related neural activity using recurrent neural newtorks (RNNs) and anatomically-accurate musculoskeletal systems. This repository contains various modules to train musculoskeletal systems using deep reinforcement learning (DRL) and RNNs in order to capture the underlying dynamics executed by the motor cortex to generate movement.

Check out the Installation to install necessary requirements.

Check out the Usage for setting up, training, and evaluating the resulting DRL algorithm.

Note

This project is under active development. If you need any help with using muSim for your application, please report in the GitHub Issues section.

Contents

• Installation
  • If facing errors
• Usage
• Basic Usage for Monkey Cycling Task
• Task Information
  • Kinematics:
    • Experimental/Task Kinematics
  • Kinematic Preprocessing:
  • Sensory Feedback Preprocessing:
• Experimental Data
  • Neural Data (optional)
  • Stimulus Data (optional)
  • Initial Pose Data (optional)
• Musculoskeletal Details
  • Musculoskeletal Model
• Biological Constraints
  • Neural Regularizations
  • Constraints on Muscle Effort Utilized
• Training Specifications
• Movement Feasibility using Inverse Kinematics and Evolutionary Algorithms
  • Inverse Kinematics
• Reward Function Specification (Optional)
• muSim Training
  • Training the uSim Controller using DRL
• Testing the muSim controller
• Post-Training Qualitative Modules
  • Kinematics Visualization
  • PCA
  • Fixed-Point Analysis
  • Rotational Dynamics
• Post-Training Quantitative Modules
  • Canonical Correlation Analysis
  • Linear Regression Analysis
  • Procrustes Analysis
• Perturbation Modules
  • Selective Feedback Elimination (SFE)
  • Sensory Perturbation
  • Neural Perturbation
  • Change Musculoskeletal Properties