1+ import sys
2+ import math
3+ import numpy as np
4+ import gymnasium as gym
5+ from gymnasium import spaces
6+ # add dynamic link library or shared object to python env
7+ sys .path .append ('/path/to/SPHinXsys/case/lib/dynamic link library or shared object' )
8+ import test_2d_owsc_python as test_2d
9+
10+
11+ class OWSCEnv (gym .Env ):
12+ """Custom Environment without rendering."""
13+ # metadata = {"render_modes": ["human", "rgb_array"], "render_fps": 30}
14+
15+ def __init__ (self , render_mode = None , parallel_envs = 0 ):
16+ # Initialize environment parameters
17+ self .parallel_envs = parallel_envs # Identifier for parallel simulation environments
18+ self .episode = 1 # Current episode number
19+ self .time_per_action = 0.1 # Time interval per action step
20+ self .low_action = - 1.0 # Minimum action value
21+ self .max_action = 1.0 # Maximum action value
22+ self .update_per_action = 10 # The action's effect is applied in smaller iterations within one action time step
23+ self .low_obs = - 10.0 # Minimum observation value
24+ self .high_obs = 10.0 # Maximum observation value
25+ self .obs_numbers = 16 # Number of observation variables
26+
27+ # Define action and observation spaces for Gym
28+ low_action = np .array ([self .low_action ]).astype (np .float32 )
29+ high_action = np .array ([self .max_action ]).astype (np .float32 )
30+ low_obs = np .full (self .obs_numbers , self .low_obs ).astype (np .float32 )
31+ high_obs = np .full (self .obs_numbers , self .high_obs ).astype (np .float32 )
32+
33+ self .action_space = spaces .Box (low_action , high_action ) # Continuous action space
34+ self .observation_space = spaces .Box (low_obs , high_obs ) # Continuous observation space
35+
36+ # Reset the environment at the beginning of each episode
37+ def reset (self , seed = None , options = None ):
38+ super ().reset (seed = seed )
39+
40+ # Initialize the OWSC simulation with the given episode and environment setup
41+ self .owsc = test_2d .owsc_from_sph_cpp (self .parallel_envs , self .episode )
42+ self .action_time_steps = 0 # Track the number of action steps
43+ self .action_time = 0.5 # Initialize action time
44+ self .damping_coefficient = 50 # Set damping coefficient for the environment
45+ self .total_reward_per_episode = 0.0 # Track total reward in each episode
46+
47+ # Start the simulation with the given action time and damping coefficient
48+ self .owsc .run_case (self .action_time , self .damping_coefficient )
49+
50+ # Initialize observation array with zero values
51+ self .observation = np .zeros (self .obs_numbers )
52+ # Fill the observation array with values from the OWSC simulation
53+ for i in range (0 , 2 ):
54+ self .observation [i ] = self .owsc .get_wave_height (i )
55+ self .observation [i + 2 ] = self .owsc .get_wave_velocity (i , 0 )
56+ self .observation [i + 4 ] = self .owsc .get_wave_velocity (i , 1 )
57+ self .observation [i + 6 ] = self .owsc .get_wave_velocity_on_flap (i , 0 )
58+ self .observation [i + 8 ] = self .owsc .get_wave_velocity_on_flap (i , 1 )
59+ self .observation [i + 10 ] = self .owsc .get_flap_position (i , 0 )
60+ self .observation [i + 12 ] = self .owsc .get_flap_position (i , 1 )
61+ self .observation [14 ] = self .owsc .get_flap_angle ()
62+ self .observation [15 ] = self .owsc .get_flap_angle_rate ()
63+
64+ self ._get_obs = self .observation .astype (np .float32 )
65+
66+ return self ._get_obs , {}
67+
68+ def step (self , action ):
69+ self .action_time_steps += 1
70+ # Apply the action to change the damping coefficient
71+ self .damping_change = 5.0 * action [0 ]
72+ # Penalty for invalid actions
73+ penality_0 = 0.0
74+ # Ensure the damping coefficient stays within valid bounds
75+ if self .damping_coefficient + self .damping_change < 0.01 :
76+ self .damping_change = 0.01 - self .damping_coefficient
77+ penality_0 = - 1.0
78+ if self .damping_coefficient + self .damping_change > 100 :
79+ self .damping_change = 100 - self .damping_coefficient
80+ penality_0 = - 1.0
81+
82+ reward_0 = 0.0
83+ for i in range (self .update_per_action ):
84+ self .flap_angle_rate_previous = self .owsc .get_flap_angle_rate ()
85+ self .damping_coefficient += self .damping_change / self .update_per_action
86+ self .action_time += self .time_per_action / self .update_per_action
87+ self .owsc .run_case (self .action_time , self .damping_coefficient )
88+ self .flap_angle_rate_now = self .owsc .get_flap_angle_rate ()
89+ # Calculate reward based on energy (flap angle rate)
90+ reward_0 += self .damping_coefficient * math .pow (0.5 * (self .flap_angle_rate_now + self .flap_angle_rate_previous ), 2 ) * self .time_per_action / self .update_per_action
91+ # Add any penalties to the reward
92+ reward = reward_0 + penality_0
93+ self .total_reward_per_episode += reward
94+
95+ # Update observations from the OWSC simulation
96+ for i in range (0 , 2 ):
97+ self .observation [i ] = self .owsc .get_wave_height (i )
98+ self .observation [i + 2 ] = self .owsc .get_wave_velocity (i , 0 )
99+ self .observation [i + 4 ] = self .owsc .get_wave_velocity (i , 1 )
100+ self .observation [i + 6 ] = self .owsc .get_wave_velocity_on_flap (i , 0 )
101+ self .observation [i + 8 ] = self .owsc .get_wave_velocity_on_flap (i , 1 )
102+ self .observation [i + 10 ] = self .owsc .get_flap_position (i , 0 )
103+ self .observation [i + 12 ] = self .owsc .get_flap_position (i , 1 )
104+ self .observation [14 ] = self .owsc .get_flap_angle ()
105+ self .observation [15 ] = self .owsc .get_flap_angle_rate ()
106+
107+ self ._get_obs = self .observation .astype (np .float32 )
108+
109+ # Log action and reward information to files
110+ with open (f'action_env{ self .parallel_envs } { self .episode } , 'a' ) as file :
111+ file .write (f'action_time: { self .action_time } { self .damping_coefficient } \n ' )
112+
113+ with open (f'reward_env{ self .parallel_envs } { self .episode } , 'a' ) as file :
114+ file .write (f'action_time: { self .action_time } { reward } \n ' )
115+
116+ # Check if the episode is done after 200 steps
117+ if self .action_time_steps > 99 :
118+ done = True
119+ with open (f'reward_env{ self .parallel_envs } , 'a' ) as file :
120+ file .write (f'episode: { self .episode } { self .total_reward_per_episode } \n ' )
121+ self .episode += 1
122+ else :
123+ done = False
124+
125+ # Return the updated observation, reward, done flag, and additional info
126+ return self ._get_obs , reward , done , False , {}
127+
128+ # Render method (optional, no rendering in this case)
129+ def render (self ):
130+ return 0
131+
132+ # Additional render frame logic (not implemented)
133+ def _render_frame (self ):
134+ return 0
135+
136+ # Close the environment and cleanup (optional)
137+ def close (self ):
138+ return 0
0 commit comments