Merge main into development

.github/workflows/tests.yaml

@@ -91,21 +91,21 @@ jobs:
         # Miniconda is available in $CONDA env var
         $CONDA/bin/conda create -n testenv --yes pip wheel gxx_linux-64 gcc_linux-64 python=${{ matrix.python-version }}
         $CONDA/envs/testenv/bin/python3 -m pip install --upgrade pip
-        $CONDA/envs/testenv/bin/pip3 install -e .[dev,dm_control]
+        $CONDA/envs/testenv/bin/pip3 install -e ".[dev,box2d,brax,dm_control,mario]"
 
     - name: Source install
       if: matrix.kind == 'source'
       run: |
         python -m pip install --upgrade pip
-        pip install -e .[dev,dm_control]
+        pip install -e ".[dev,box2d,brax,dm_control,mario]"
 
     - name: Dist install
       if: matrix.kind == 'dist'
       run: |
         python -m pip install --upgrade pip
         python setup.py sdist
         last_dist=$(ls -t dist/carl-*.tar.gz | head -n 1)
-        pip install $last_dist[dev,dm_control]
+        pip install $last_dist[dev,box2d,brax,dm_control,mario]
 
     - name: Tests
       timeout-minutes: 60

.gitignore

@@ -23,9 +23,7 @@ outputs
 testvenv
 *.egg-info
 runs
-*.png
 *.pdf
-*.csv
 *.pickle
 *.ipynb_checkpoints
 *optgap*

README.md

@@ -54,18 +54,8 @@ RNA and Mario at this point, however, will not run on any operation system besid
 
 To install ToadGAN for the Mario environment:
 ```bash
-git submodule update --init --recursive
-
-# if this does not work, clone manually
-git clone https://github.com/frederikschubert/Mario-AI-Framework carl/envs/mario/Mario-AI-Framework
-git clone https://github.com/Mawiszus/TOAD-GUI  carl/envs/mario/TOAD-GUI
-
 # System requirements
 sudo apt install libfreetype6-dev xvfb
-
-# Compile java source files
-cd carl/envs/mario/Mario-AI-Framework/src
-javac *.java
 ```
 
 If you want to use RNA, please take a look at the associated [ReadME](carl/envs/rna/readme.md).

carl/__init__.py

@@ -6,7 +6,7 @@
 import datetime
 
 name = "CARL"
-package_name = "carl"
+package_name = "carl-bench"
 author = __author__
 
 author_email = "[email protected]"

carl/context/context_space.py

@@ -2,6 +2,8 @@
 
 from typing import List
 
+import warnings
+
 import gymnasium.spaces as spaces
 import numpy as np
 from ConfigSpace.hyperparameters import (
@@ -16,6 +18,8 @@
 
 from carl.utils.types import Context, Contexts
 
+warnings.filterwarnings("ignore", message="invalid value encountered in scalar divide")
+
 ContextFeature: TypeAlias = Hyperparameter
 NumericalContextFeature: TypeAlias = NumericalHyperparameter
 NormalFloatContextFeature: TypeAlias = NormalFloatHyperparameter

carl/envs/__init__.py

@@ -48,6 +48,6 @@ def check_spec(spec_name: str) -> bool:
 if found:
     from carl.envs.dmc import *
 
-found = check_spec("distance")
-if found:
-    from carl.envs.rna import *
+# found = check_spec("distance")
+# if found:
+#     from carl.envs.rna import *

carl/envs/brax/carl_brax_env.py

@@ -251,4 +251,4 @@ def _update_context(self) -> None:
                 updated_geoms.append(geom_new)
             sys = sys.replace(geoms=updated_geoms)
 
-        self.env.sys = sys
+        self.env.unwrapped.sys = sys

carl/envs/gymnasium/box2d/carl_bipedal_walker.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import numpy as np
 from Box2D.b2 import edgeShape, fixtureDef, polygonShape
 from gymnasium.envs.box2d import bipedal_walker
 from gymnasium.envs.box2d import bipedal_walker as bpw
@@ -87,6 +86,9 @@ def get_context_features() -> dict[str, ContextFeature]:
 
     def _update_context(self) -> None:
         self.env: bipedal_walker.BipedalWalker
+        self.context = CARLBipedalWalker.get_context_space().insert_defaults(
+            self.context
+        )
         bpw.FPS = self.context["FPS"]
         bpw.SCALE = self.context["SCALE"]
         bpw.FRICTION = self.context["FRICTION"]
@@ -112,14 +114,14 @@ def _update_context(self) -> None:
         gravity_y = self.context["GRAVITY_Y"]
 
         gravity = (gravity_x, gravity_y)
-        self.env.world.gravity = gravity
+        self.env.unwrapped.world.gravity = gravity
 
         # Important for building terrain
-        self.env.fd_polygon = fixtureDef(
+        self.env.unwrapped.fd_polygon = fixtureDef(
             shape=polygonShape(vertices=[(0, 0), (1, 0), (1, -1), (0, -1)]),
             friction=bipedal_walker.FRICTION,
         )
-        self.env.fd_edge = fixtureDef(
+        self.env.unwrapped.fd_edge = fixtureDef(
             shape=edgeShape(vertices=[(0, 0), (1, 1)]),
             friction=bipedal_walker.FRICTION,
             categoryBits=0x0001,
@@ -152,95 +154,4 @@ def _update_context(self) -> None:
             maskBits=0x001,
         )
 
-        self.env.world.gravity = gravity
-
-
-def demo_heuristic(env: CARLBipedalWalker | bipedal_walker.BipedalWalker) -> None:
-    env.reset()
-    steps = 0
-    total_reward = 0
-    a = np.array([0.0, 0.0, 0.0, 0.0])
-    STAY_ON_ONE_LEG, PUT_OTHER_DOWN, PUSH_OFF = 1, 2, 3
-    SPEED = 0.29  # Will fall forward on higher speed
-    state = STAY_ON_ONE_LEG
-    moving_leg = 0
-    supporting_leg = 1 - moving_leg
-    SUPPORT_KNEE_ANGLE = +0.1
-    supporting_knee_angle = SUPPORT_KNEE_ANGLE
-    while True:
-        s, r, terminated, truncated, info = env.step(a)
-        s = s["state"]
-        total_reward += r
-        if steps % 20 == 0 or terminated or truncated:
-            print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
-            print("step {} total_reward {:+0.2f}".format(steps, total_reward))
-            print("hull " + str(["{:+0.2f}".format(x) for x in s[0:4]]))
-            print("leg0 " + str(["{:+0.2f}".format(x) for x in s[4:9]]))
-            print("leg1 " + str(["{:+0.2f}".format(x) for x in s[9:14]]))
-        steps += 1
-
-        contact0 = s[8]  # noqa: F841
-        contact1 = s[13]  # noqa: F841
-        moving_s_base = 4 + 5 * moving_leg
-        supporting_s_base = 4 + 5 * supporting_leg
-
-        hip_targ = np.array([None, None])  # -0.8 .. +1.1
-        knee_targ = np.array([None, None])  # -0.6 .. +0.9
-        hip_todo = np.array([0.0, 0.0])
-        knee_todo = np.array([0.0, 0.0])
-
-        if state == STAY_ON_ONE_LEG:
-            hip_targ[moving_leg] = 1.1
-            knee_targ[moving_leg] = -0.6
-            supporting_knee_angle += 0.03
-            if s[2] > SPEED:
-                supporting_knee_angle += 0.03
-            supporting_knee_angle = min(supporting_knee_angle, SUPPORT_KNEE_ANGLE)
-            knee_targ[supporting_leg] = supporting_knee_angle
-            if s[supporting_s_base + 0] < 0.10:  # supporting leg is behind
-                state = PUT_OTHER_DOWN
-        if state == PUT_OTHER_DOWN:
-            hip_targ[moving_leg] = +0.1
-            knee_targ[moving_leg] = SUPPORT_KNEE_ANGLE
-            knee_targ[supporting_leg] = supporting_knee_angle
-            if s[moving_s_base + 4]:
-                state = PUSH_OFF
-                supporting_knee_angle = min(s[moving_s_base + 2], SUPPORT_KNEE_ANGLE)
-        if state == PUSH_OFF:
-            knee_targ[moving_leg] = supporting_knee_angle
-            knee_targ[supporting_leg] = +1.0
-            if s[supporting_s_base + 2] > 0.88 or s[2] > 1.2 * SPEED:
-                state = STAY_ON_ONE_LEG
-                moving_leg = 1 - moving_leg
-                supporting_leg = 1 - moving_leg
-
-        if hip_targ[0]:
-            hip_todo[0] = 0.9 * (hip_targ[0] - s[4]) - 0.25 * s[5]
-        if hip_targ[1]:
-            hip_todo[1] = 0.9 * (hip_targ[1] - s[9]) - 0.25 * s[10]
-        if knee_targ[0]:
-            knee_todo[0] = 4.0 * (knee_targ[0] - s[6]) - 0.25 * s[7]
-        if knee_targ[1]:
-            knee_todo[1] = 4.0 * (knee_targ[1] - s[11]) - 0.25 * s[12]
-
-        hip_todo[0] -= 0.9 * (0 - s[0]) - 1.5 * s[1]  # PID to keep head strait
-        hip_todo[1] -= 0.9 * (0 - s[0]) - 1.5 * s[1]
-        knee_todo[0] -= 15.0 * s[3]  # vertical speed, to damp oscillations
-        knee_todo[1] -= 15.0 * s[3]
-
-        a[0] = hip_todo[0]
-        a[1] = knee_todo[0]
-        a[2] = hip_todo[1]
-        a[3] = knee_todo[1]
-        a = np.clip(0.5 * a, -1.0, 1.0)
-
-        env.render()
-        if terminated or truncated:
-            break
-
-
-if __name__ == "__main__":
-    # Heurisic: suboptimal, have no notion of balance.
-    env = CARLBipedalWalker()
-    demo_heuristic(env)
-    env.close()
+        self.env.unwrapped.world.gravity = gravity

carl/envs/gymnasium/box2d/carl_lunarlander.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from Box2D.b2 import vec2
 from gymnasium.envs.box2d import lunar_lander
 from gymnasium.envs.box2d.lunar_lander import LunarLander
 
@@ -77,8 +78,12 @@ def _update_context(self) -> None:
             if hasattr(lunar_lander, key):
                 setattr(lunar_lander, key, value)
 
-        gravity_x = self.context["GRAVITY_X"]
-        gravity_y = self.context["GRAVITY_Y"]
+        gravity_x = self.context.get(
+            "GRAVITY_X", self.get_context_features()["GRAVITY_X"].default_value
+        )
+        gravity_y = self.context.get(
+            "GRAVITY_Y", self.get_context_features()["GRAVITY_Y"].default_value
+        )
 
-        gravity = (gravity_x, gravity_y)
-        self.env.world.gravity = gravity
+        gravity = vec2(float(gravity_x), float(gravity_y))
+        self.env.unwrapped.world.gravity = gravity