First Simulation

The Quick Start introduces world construction and one environment step. This walkthrough connects those pieces into a complete rollout workflow: choose actions, advance the simulator, check the result, and save an animation.

Use the source checkout with the visualization extra installed. Run commands from the repository root:

python -m pip install -e ".[viz]"
python -m examples.first_motion --steps 300 --output /tmp/first-motion.gif

Open /tmp/first-motion.gif. You should see a three-cell body deform above flat ground. The command prints the frame count and total reward. It checks that positions are finite before saving. Choose a different output path on Windows.

This example demonstrates actuation and rendering. Its sinusoidal controller is not a trained walking policy, and forward travel is not guaranteed.

What runs

1. WalkerV0 builds a world around a body containing horizontal and vertical actuators.
2. reset() returns the initial observation and explicit environment state.
3. A periodic controller produces one multiplier per actuator, centred on 1.0.
4. jax.lax.scan applies the actions in sequence and records positions and spring lengths.
5. render_episode draws the saved trajectory; save_gif writes the animation.

The default 300 environment steps represent 0.9 seconds of simulated time: each step contains 30 physics updates of 0.0001 seconds. GIF playback settings can change the apparent speed. The first run includes JAX compilation, so it is not a useful throughput measurement.

The shared rollout helper keeps the terminal transition and freezes the state on later steps. Built-in environments otherwise keep advancing physics after done becomes true.

Read examples/first_motion.py and examples/common.py for the complete implementation. Next, load your own design or write a controller.