Scripts for "Motor Synergy Generalization Framework for New Targets in Multi-planar and Multi-directional Reaching Task"
- Python 3.8.5
- cmaes: 0.8.2
- gym: 0.15.7
- matplotlib: 3.4.2
- mujoco-py: 2.0.2.1
- numpy: 1.20.3
- scikit-learn: 0.24.2
- scipy: 1.6.3
- spinup: 0.2.0 (Available at https://github.com/openai/spinningup.git)
- torch: 1.9.0
- MuJoCo version 2.0
python src/env_test/myenv.py
Horizontal policy:
python src/rl/train.py --out results/result_h --task horizontal --gpu 2 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Sagittal policy:
python src/rl/train.py --out results/result_v --task vertical --gpu 3 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Assume that there exists a folder named result.
Eight-points reaching in the learned targets:
python src/rl/evaluate.py --model results/result_h/pyt_save/model.pt --norender --out result/result_h.pickle
python src/rl/evaluate.py --model results/result_v/pyt_save/model.pt --norender --out result/result_s.pickle
32-points reaching in the learned planes:
python src/rl/evaluate.py --model results/result_h/pyt_save/model.pt --n-rollouts 32 --norender --out result/result_h-32p.pickle
python src/rl/evaluate.py --model results/result_v/pyt_save/model.pt --n-rollouts 32 --norender --out result/result_s-32p.pickle
Eight-points reaching in the frontal targets:
python src/rl/evaluate.py --model results/result_h/pyt_save/model.pt --task facing --norender --out result/result_h-frontal.pickle
python src/rl/evaluate.py --model results/result_v/pyt_save/model.pt --task facing --norender --out result/result_s-frontal.pickle
Eight-points reaching in the upper targets:
python src/rl/evaluate.py --model results/result_h/pyt_save/model.pt --task upper --norender --out result/result_h-upper.pickle
python src/rl/evaluate.py --model results/result_v/pyt_save/model.pt --task upper --norender --out result/result_s-upper.pickle
Trajectories of eight-points reaching in the learned targets:
python src/rl/show_results.py --filename result/result_h.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-horizontal.pdf
python src/rl/show_results.py --filename result/result_s.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-sagittal.pdf
Trajectories of 32-points reaching in the learned planes:
python src/rl/show_results.py --filename result/result_h-32p.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-horizontal-32p.pdf
python src/rl/show_results.py --filename result/result_s-32p.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-sagittal-32p.pdf
Trajectories of eight-points reaching in the frontal targets:
python src/rl/show_results.py --filename result/result_h-frontal.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-frontal-by-horizontal-policy.pdf
python src/rl/show_results.py --filename result/result_s-frontal.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-frontal-by-sagittal-policy.pdf
Trajectories of eight-points reaching in the upper targets:
python src/rl/show_results.py --filename result/result_h-upper.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-upper-by-horizontal-policy.pdf
python src/rl/show_results.py --filename result/result_s-upper.pickle -v pos2d err
mv result/trajectories_original.pdf result/trajectories-upper-by-sagittal-policy.pdf
Training:
python src/rl/train.py --out results/result_h_save --task horizontal --gpu 2 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
python src/rl/train.py --out results/result_v_save --task vertical --gpu 3 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Evaluation:
./src/rl/generate_progress.sh
Visualization:
python src/rl/show_traj_progress.py
python src/synergetic/show_R2.py --filename result/result_h.pickle
mv result/R2.pdf result/R2-horizontal.pdf
python src/synergetic/show_R2.py --filename result/result_s.pickle
mv result/R2.pdf result/R2-sagittal.pdf
Motor synergy waveforms:
python src/synergetic/show_synergy.py --filename result/result_h.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-horizontal.pdf
python src/synergetic/show_synergy.py --filename result/result_s.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-sagittal.pdf
Motor synergy activities for reaching directions:
python src/synergetic/show_synergy_direction.py --filename result/result_h.pickle --filename2 result/result_h-32p.pickle --n-synergies 4
mv result/synergy_direction.pdf result/synergy_direction-horizontal.pdf
python src/synergetic/show_synergy_direction.py --filename result/result_s.pickle --filename2 result/result_s-32p.pickle --n-synergies 4
mv result/synergy_direction.pdf result/synergy_direction-sagittal.pdf
./src/synergetic/optimization_all.sh
Performance maps:
python src/synergetic/show_performance_map.py --targets horizontal
mv result/performance_map.pdf result/performance_map-horizontal.pdf
python src/synergetic/show_performance_map.py --targets sagittal
mv result/performance_map.pdf result/performance_map-sagittal.pdf
python src/synergetic/show_performance_map.py --targets frontal
mv result/performance_map.pdf result/performance_map-frontal.pdf
python src/synergetic/show_performance_map.py --targets upper
mv result/performance_map.pdf result/performance_map-upper.pdf
Trajectories in the frontal targets:
python src/synergetic/show_results.py --filename result/optim_f1/result_5_4/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-frontal-by-optimization.pdf
python src/synergetic/show_results.py --filename result/optim_f1/result_5_0/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-frontal-by-optimization-h.pdf
python src/synergetic/show_results.py --filename result/optim_f1/result_0_8/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-frontal-by-optimization-v.pdf
python src/synergetic/show_traj_progress.py --filename result/optim_f1/result_5_4/intermediate_results.pickle
mv result/trajectory_progress.pdf result/trajectory_progress-frontal.pdf
Trajectories in the upper targets:
python src/synergetic/show_results.py --filename result/optim_u1/result_5_5/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-upper-by-optimization.pdf
python src/synergetic/show_results.py --filename result/optim_u1/result_8_0/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-upper-by-optimization-h.pdf
python src/synergetic/show_results.py --filename result/optim_u1/result_0_6/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-upper-by-optimization-v.pdf
python src/synergetic/show_traj_progress.py --filename result/optim_u1/result_5_5/intermediate_results.pickle
mv result/trajectory_progress.pdf result/trajectory_progress-upper.pdf
Trajectories in the horizontal targets:
python src/synergetic/show_results.py --filename result/optim_h1/result_5_0/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-horizontal-by-optimization.pdf
Trajectories in the sagittal targets:
python src/synergetic/show_results.py --filename result/optim_s1/result_0_5/optimization_results.pickle -v pos2d err
mv result/trajectories_synergetic.pdf result/trajectories-sagittal-by-optimization.pdf
Optimization progress:
python src/synergetic/show_traj_progress.py --filename result/optim_f1/result_5_4/optimization_results.pickle
mv result/trajectory_progress.pdf result/trajectory_progress-frontal.pdf
Horizontal policy for the shorter-arm:
python src/rl/train.py --out results/result_h_short --task horizontal --arm short --gpu 0 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Sagittal policy for the shorter-arm:
python src/rl/train.py --out results/result_s_short --task vertical --arm short --gpu 0 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Horizontal policy for the longer-arm:
python src/rl/train.py --out results/result_h_long --task horizontal --arm long --gpu 0 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Sagittal policy for the longer-arm:
python src/rl/train.py --out results/result_s_long --task vertical --arm long --gpu 0 --epochs 3000 --k1 0.002 --k2 0.2 --activation relu --hidden-units 256 256
Assume that there exists a folder named result.
Eight-points reaching in the learned targets:
python src/rl/evaluate.py --model results/result_h_short/pyt_save/model.pt --norender --out result/result_h_short.pickle
python src/rl/evaluate.py --model results/result_s_short/pyt_save/model.pt --norender --out result/result_s_short.pickle
python src/rl/evaluate.py --model results/result_h_long/pyt_save/model.pt --norender --out result/result_h_long.pickle
python src/rl/evaluate.py --model results/result_s_long/pyt_save/model.pt --norender --out result/result_s_long.pickle
Motor synergy waveforms:
python src/synergetic/show_synergy.py --filename result/result_h_short.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-horizontal_short.pdf
python src/synergetic/show_synergy.py --filename result/result_s_short.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-sagittal_short.pdf
python src/synergetic/show_synergy.py --filename result/result_h_long.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-horizontal_long.pdf
python src/synergetic/show_synergy.py --filename result/result_s_long.pickle --n-synergies 4
mv result/motor_synergies.pdf result/motor_synergies-sagittal_long.pdf
./src/synergetic/optimization_all2.sh
Performance maps:
python src/synergetic/show_performance_map2.py --targets frontal --arm short
mv result/performance_map.pdf result/performance_map-frontal-short.pdf
python src/synergetic/show_performance_map2.py --targets frontal --arm long
mv result/performance_map.pdf result/performance_map-frontal-long.pdf