Publish repository: Thu Nov 27 15:25:23 UTC 2025

Author: github-actions[bot]

src/exercise01/exercise01/exercise01.ipynb

@@ -254,7 +254,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "The following is quite some text, but it allows you to gain a background understanding of the framework we use.\n",
+    "The following helps you to gain a background understanding of the framework we use.\n",
     "\n",
     "### 2.1 Background\n",
     "\n",
@@ -432,35 +432,6 @@
     "> **Note**: The actions need to be a 2D array of shape `(n_envs, n_actions)`. Except for parallel simulations on GPU for learning based methods, we set `n_envs=1`."
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "##### Simulation\n",
-    "\n",
-    "An efficient, highly-parallizable (GPU) simulation of the discretized dynamics models is implemented in our simulator [`crazyflow`](https://github.com/utiasDSL/crazyflow).\n",
-    "\n",
-    "- The simulation is implemented in [JAX](https://docs.jax.dev/en/latest/quickstart.html), which is basically numpy for the GPU, and enables us to massively parallize the simulation.\n",
-    "- The simulation provides physics, integrators (Euler or RK4), lower level controllers, and visualisation routines. If you are interested, you can check out the [this main function](https://github.com/utiasDSL/crazyflow/blob/d87bc1eaf100e7d8927731c630e52a7163108ecf/crazyflow/sim/sim.py) of the simulation.\n",
-    "- The default simulation step-size is $2\\text{ms}$. The simulation provides Explicit Euler or RK4 for integration. We recommend to stick with the values that have been set for this exercise.\n",
-    "- For visualisation, we use [MuJoCo](https://mujoco.org/), or more specific [MJX](https://mujoco.readthedocs.io/en/stable/mjx.html). While MuJoCo also comes with its own physics engine, we only use it for vizualization in crazyflow.\n",
-    "\n",
-    "> **Note**: Crazyflow is of course already installed in the container. You can find the code in `/home/vscode/venv/lib/python3.11/site-packages/crazyflow`, but **do not make any changes**. For the exercises, we use [`crazyflow v0.0.2`](https://github.com/utiasDSL/crazyflow/tree/v0.0.2).\n",
-    "\n",
-    "\n",
-    "\n",
-    "##### Gymnasium environment\n",
-    "\n",
-    "The simulation is wrapped with different [Gymnasium environments](https://github.com/utiasDSL/crazyflow/blob/d87bc1eaf100e7d8927731c630e52a7163108ecf/crazyflow/gymnasium_envs/crazyflow.py) that define different tasks. You will use the environments to interact with the drone simulation using your controllers, as explained in the previous section. We provide [multiple environments](https://github.com/utiasDSL/crazyflow/blob/d87bc1eaf100e7d8927731c630e52a7163108ecf/crazyflow/gymnasium_envs/crazyflow.py), each for a different purpose. The environments differ in their goals (reward formulations), observation spaces, and other details. For instance, we have `CrazyflowEnvReachGoal` and `CrazyflowEnvFigureEightTrajectory`. We will use the different environments throughout the exercises.\n",
-    "- **Observation space**: The observation spaces differ for the different gymnasium environments. However, each observation includes at least the state of the drone, which are `position`, `orientation` (in quaternion), and its derivatives. Some environments include additional observations, e.g. the `CrazyflowEnvReachGoal` additionally includes the `difference_to_goal`.\n",
-    "- **Action space** (corresponds to control input $u$): We use the [Attitude controll interface](https://www.bitcraze.io/documentation/repository/crazyflie-firmware/master/functional-areas/sensor-to-control/controllers/) of the drone, as specified in the dynamics model above. This interface takes **four control inputs**: `[collective thrust, roll, pitch, yaw]`. Thus, the action space is a [continuous box](https://gymnasium.farama.org/api/spaces/fundamental/#gymnasium.spaces.Box) represented as <b>float32</b>. The range for the thrust is <b>0.11264675 to 0.5933658</b>, while the orientations range from <b>-π/2 to π/2</b>. Actions needs to be a 2D array of shape `(n_envs, n_actions)`. For all exercises except the last one, we always have `n_envs=1` and `n_actions=4`.\n",
-    "\n",
-    "\n",
-    "> **Note**: There are different ways to represent orientations in 3D space. The observations of the environment returns [quaternions](https://en.wikipedia.org/wiki/Quaternions_and_spatial_rotation), while the symbolic model works with Euler Angles. You do not have to worry too much about that, as you will use [this scipy package](https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.transform.Rotation.html) to easily convert orientation representations.\n",
-    "\n",
-    "> **Note**: The actions need to be a 2D array of shape `(n_envs, n_actions)`. Except for parallel simulations on GPU for learning based methods, we set `n_envs=1`."
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},