Emergent Locomotion in HSA Robot Documentation

This project focused on modeling, simulating, and developing locomotion gaits for a soft-robotic module based on Handed Shearing Auxetic (HSA) actuators. The primary goal was to reproduce the physical module’s behavior in simulation, and then apply learning-based methods to discover novel and unintuitive gaits. This documentation contains the definition of the HSA Gym environment, the curriculum learning framework, terrain generation utilities, and reinforcement learning training and evaluation scripts. Additionally, it includes documentation for the MuJoCo simulation API used for preliminary testing and demonstration of control strategies.

Contents:

• HSA Environment Modules
• MuJoCo Base RL Class
  • CustomMujocoEnv
  • expand_model_path()
• HSA Gym Environment and Utilities
  • HSAEnv
  • GoalCurriculumManager
• Training Script
  • CurriculumWrapper
  • RewardComponentLogger
  • SaveCurriculumCallback
  • SaveVecNormalizeCallback
  • extract_step_number()
  • get_latest_checkpoint()
  • get_latest_curriculum()
  • get_latest_vecnormalize()
  • load_config()
  • main()
  • make_curriculum_env()
• Evaluation Script
  • analyze_actions()
  • extract_step_number()
  • find_matching_vecnormalize()
  • load_config()
  • make_env()
• Terrain Generation Utilities
  • ensure_flat_spawn_zone()
  • generate_corridor_terrain()
  • generate_crater_terrain()
  • generate_flat_terrain()
  • generate_flat_with_incline()
  • generate_flat_with_ramp()
  • generate_incline_terrain()
  • generate_perlin_noise()
  • generate_poles_terrain()
  • generate_spiral_track()
  • generate_stairs_terrain()
  • generate_terrain()
  • generate_wavy_field()
• MuJoCo Control Interface