Add ls20 env

config/ls20.ini

@@ -0,0 +1,22 @@
+[base]
+env_name = ls20
+
+[env]
+fps = 30
+reset_enabled = 0
+
+[vec]
+total_agents = 1024
+num_buffers = 2
+num_threads = 4
+
+[policy]
+hidden_size = 256
+num_layers = 2
+
+[train]
+total_timesteps = 100_000_000
+gamma = 0.99
+minibatch_size = 16384
+horizon = 128
+use_rnn = 1

ocean/ls20/binding.c

@@ -0,0 +1,25 @@
+#include "ls20.h"
+
+#define OBS_SIZE LS20_OBS_SIZE
+#define NUM_ATNS 1
+#define ACT_SIZES {LS20_ACTION_COUNT}
+#define OBS_TENSOR_T ByteTensor
+#define MY_ACTION_MASK LS20_ACTION_COUNT
+
+#define Env Ls20
+#include "vecenv.h"
+
+void my_init(Env* env, Dict* kwargs) {
+    env->num_agents = 1;
+    DictItem* fps = dict_get_unsafe(kwargs, "fps");
+    env->fps = fps == NULL ? LS20_DEFAULT_FPS : (int)fps->value;
+    DictItem* reset_enabled = dict_get_unsafe(kwargs, "reset_enabled");
+    env->reset_enabled = reset_enabled != NULL && reset_enabled->value != 0.0;
+}
+
+void my_log(Log* log, Dict* out) {
+    dict_set(out, "perf", log->perf);
+    dict_set(out, "score", log->score);
+    dict_set(out, "episode_return", log->episode_return);
+    dict_set(out, "episode_length", log->episode_length);
+}

ocean/ls20/ls20.c

@@ -0,0 +1,84 @@
+#include "ls20.h"
+
+static const char* WINNING_LEVELS[LS20_LEVEL_COUNT] = {
+    "LLLUUUURRRUUU",
+    "URUUUUURRDRDDDDDDDLLRURLRUUUUUUULLLLLLDLDDDDD",
+    "UUUUUUUULDDDDDDDDUUULLUUURRRDDRRRRUUULRLDRDDDDDDD",
+    "LLLDDDLLLRRUUULLUDDUDUDUDUDUDUDUULLUDDULLUUUDLULUUUURURUULLL",
+    "ULUULDLULRLRLRUDLRLDDLLLULRLRDLDDDDDRDURRULRDDDRRRRRRDUU",
+    ("UUDULRRRRUUULLRRDDDDLLLURRRUUUUUUUDDRRUURDDDUUULUDLLLRLRLLLL"
+     "DDDRRRDULDLLUUUUUUURLRRRRLLLULDDDDDDDRRDURRRRUUUUURRUURDDDDD"),
+    "LLDDDDDRLRLRRLDURLRLRLRLDUULUURRRRUDRRUURRUUUUUURDLDDDDDLLUDDDD",
+};
+
+static int action_from_char(char action) {
+    if (action == 'U') return ACTION1;
+    if (action == 'D') return ACTION2;
+    if (action == 'L') return ACTION3;
+    assert(action == 'R');
+    return ACTION4;
+}
+
+static void demo(void) {
+    unsigned char observations[LS20_OBS_SIZE] = {0};
+    float action = 0.0f;
+    float reward = 0.0f;
+    float terminal = 0.0f;
+    Ls20 env = {
+        .num_agents = 1,
+        .fps = LS20_DEFAULT_FPS,
+        .observations = observations,
+        .actions = &action,
+        .rewards = &reward,
+        .terminals = &terminal,
+    };
+    c_reset(&env);
+    c_render(&env);
+    for (int level = 0; level < LS20_LEVEL_COUNT; level++) {
+        for (const char* step = WINNING_LEVELS[level]; *step; step++) {
+            action = (float)action_from_char(*step);
+            c_step(&env);
+            c_render(&env);
+        }
+    }
+    c_close(&env);
+}
+
+static void play(void) {
+    unsigned char observations[LS20_OBS_SIZE] = {0};
+    float action = 0.0f;
+    float reward = 0.0f;
+    float terminal = 0.0f;
+    Ls20 env = {
+        .num_agents = 1,
+        .observations = observations,
+        .actions = &action,
+        .rewards = &reward,
+        .terminals = &terminal,
+    };
+    c_reset(&env);
+    c_render(&env);
+    while (!WindowShouldClose()) {
+        int next_action = -1;
+        if (IsKeyPressed(KEY_UP) || IsKeyPressed(KEY_W)) next_action = ACTION1;
+        if (IsKeyPressed(KEY_DOWN) || IsKeyPressed(KEY_S)) next_action = ACTION2;
+        if (IsKeyPressed(KEY_LEFT) || IsKeyPressed(KEY_A)) next_action = ACTION3;
+        if (IsKeyPressed(KEY_RIGHT) || IsKeyPressed(KEY_D)) next_action = ACTION4;
+        if (IsKeyPressed(KEY_R)) next_action = RESET;
+        if (next_action >= 0) {
+            action = (float)next_action;
+            c_step(&env);
+        }
+        c_render(&env);
+    }
+    c_close(&env);
+}
+
+int main(int argc, char** argv) {
+    if (argc == 2 && strcmp(argv[1], "--demo") == 0) {
+        demo();
+        return 0;
+    }
+    play();
+    return 0;
+}