Hexamind/iPADS

1import time2import streamlit as st3from params import *4from stable_baselines3 import SAC5import pandas as pd6import pydeck as pdk7from TwoDimEnv import TwoDimEnv8 9import gymnasium as gym10 11def run_episode(env, model, lat_tg, lon_tg, rho_init=RHO_INIT, theta_init=THETA_INIT, zed=Z_INIT):12    '''13    runs the episode given an observation init ... and an env and a model14    :param env: TwoDimEnv15    :param model:16    :param lat_tg:  lattitude de la target (0,0)17    :param lon_tg: longitude de la target (0,0)18    :param rho_init:19    :param theta_init:20    :param zed:21    :return:22    '''23    obs = env.reset(training=False, rho_init=rho_init, theta_init=theta_init, z_init=zed)24    step = 025    lat, lon = rhotheta_to_latlon(MOVE_TO_METERS * rho_init, theta_init, lat_tg, lon_tg)26    path = [[lon, lat, MOVE_TO_METERS * zed]]27    traj = [lat, lon, MOVE_TO_METERS * zed]28    while step < zed:29        step += 130        action, _ = model.predict(obs, deterministic=True)31        obs, _, _, _ = env.step(action)32        rho = obs[0] * env.space_limits33        theta = obs[1] * 2 * PI34        lat, lon = rhotheta_to_latlon(MOVE_TO_METERS*rho, theta, lat_tg, lon_tg)35        traj = np.vstack((traj, [lat, lon, MOVE_TO_METERS*(zed - step)]))36        path.append([lon, lat, MOVE_TO_METERS*(zed - step)])37    df_col = pd.DataFrame(traj, columns=['lat', 'lon', 'zed'])38    df_path = pd.DataFrame([{39        'color': [0, 0, 200, 120],40        'path': path41    }])42    return df_path, df_col43 44def rhotheta_to_latlon(rho, theta, lat_tg, lon_tg):45    '''46    transforms polar coordinates into lat, lon47    :param rho:48    :param theta:49    :param lat_tg: latitude de la target (0,0)50    :param lon_tg: longitude de la target (0,0)51    :return:52    '''53    z = rho * np.exp(1j * theta)54    lat = np.imag(z)*360/(40075*1000) + lat_tg55    lon = np.real(z)*360/(40075*1000*np.cos(PI/180*lat)) + lon_tg56    return lat, lon57 58 59def get_layers(df, df_past, df_target, df_path, df_col):60    '''61    renders the layers to be displayed with df in different formats as entries62    :param df:63    :param df_past:64    :param df_target:65    :param df_path:66    :param df_col:67    :return:68    '''69    return [70        pdk.Layer(71            'ScatterplotLayer',72            data=df,73            get_position='[lon, lat]',74            get_color='[200, 30, 0, 160, 40]',75            get_radius=10,76        ),77        pdk.Layer(78            'ScatterplotLayer',79            data=df_target,80            get_position='[lon, lat]',81            get_color='[200, 30, 0]',82            get_radius=65,83        ),84        pdk.Layer(85            'ScatterplotLayer',86            data=df_target,87            get_position='[lon, lat]',88            get_color='[255, 255, 255]',89            get_radius=45,90        ),91        pdk.Layer(92            'ScatterplotLayer',93            data=df_target,94            get_position='[lon, lat]',95            get_color='[0, 0, 200]',96            get_radius=25,97        ),98        pdk.Layer(99            'ScatterplotLayer',100            data=df_past,101            get_position='[lon, lat]',102            get_color='[200, 30, 0, 40]',103            get_radius=10,104        ),105        pdk.Layer(106            type="PathLayer",107            data=df_path,108            pickable=True,109            get_color="color",110            width_scale=20,111            width_min_pixels=1,112            get_path="path",113            get_width=1,114        ),115        pdk.Layer(116            type="ColumnLayer",117            data=df_col,118            get_position=['lon', 'lat'],119            get_elevation="zed",120            elevation_scale=1,121            radius=10,122            get_fill_color=[160, 20, 0, 10],123            pickable=True,124            auto_highlight=True125        ),126    ]127 128 129def show():130    '''131    shows the i-PADS in Streamlit132    :return:133    '''134    env = TwoDimEnv()135    model = SAC.load("longModel")136    st.title('Intelligent PADS by hexamind')137    st.write('This is a quick demo of an autonomous Parachute (Precision Air Delivery System) controlled by Reinforcement learning. ')138    st.text('<- Set the starting point')139 140    st.sidebar.write("Where do you want the parachute to start from?")141    rho = st.sidebar.slider('What distance? (in m)', 0, 3000, 1500) / MOVE_TO_METERS142    theta = 2*PI/360 * st.sidebar.slider('What angle?', 0, 360, 90)143    zed = int(st.sidebar.slider('What elevation? (in m)', 0, 1200, 600) / MOVE_TO_METERS)144 145    location = st.sidebar.radio("Location", ['San Francisco', 'Paris', 'Puilaurens'])146    lat_tg = LOC[location]['lat']147    lon_tg = LOC[location]['lon']148    df_path, df_col = run_episode(env, model, lat_tg, lon_tg, rho_init=rho, theta_init=theta, zed=zed)149    st.sidebar.write(150        'If you like to play, you will probably find some starting points where the parachute is out of control :) '151        'No worries, we have plenty more efficient models at www.hexamind.ai ')152 153    df_target = pd.DataFrame({'lat': [lat_tg], 'lon': [lon_tg]})154    deck_map = st.empty()155    pitch = st.slider('pitch', 0, 100, 50)156    initial_view_state = pdk.ViewState(157            latitude=lat_tg,158            longitude=lon_tg,159            zoom=12,160            pitch=pitch161    )162    deck_map.pydeck_chart(pdk.Deck(163        map_style='mapbox://styles/mapbox/light-v9',164        initial_view_state=initial_view_state165    ))166    df_pathi = df_path.copy()167    for i in range(zed):168        df_pathi['path'][0] = df_path['path'][0][0:i+1]169        layers = get_layers(df_col[i:i+1], df_col[0:i], df_target, df_pathi, df_col[0:i+1])170        deck_map.pydeck_chart(pdk.Deck(171            map_style='mapbox://styles/mapbox/light-v9',172            initial_view_state=initial_view_state,173            layers=layers174        ))175        time.sleep(TIMESLEEP)176 177 178show()179 180# to uncomment for debug181#show_print()182