Visualizer

PufferDrive uses Raylib for rendering the environment. Rendering is driven from Python using the torch policy directly. No separate binary or weight export is required.

Dependencies

For headless rendering, we need ffmpeg and xvfb.

sudo apt update && sudo apt install ffmpeg xvfb

Render Modes

Configure render_mode in pufferlib/config/ocean/drive.ini:

; 0 = pop-up window (requires display)
; 1 = headless (pipes frames to ffmpeg, recommended for servers/training)
render_mode = 1

Rendering once

puffer eval puffer_drive

This runs a short rollout, calls env.render() each step, and finalizes the video on vecenv.close(). Use render_mode to determine whether the video shows up as a pop-up window, or whether it is stored as an mp4.

View modes

Control what is rendered via the view_mode argument to env.render():

class RenderView(IntEnum):
    FULL_SIM_STATE = 0  # Top-down, fully observable
    BEV_AGENT_OBS  = 1  # Top-down, selected agent's observations only
    AGENT_PERSP    = 2  # Third-person perspective following selected agent

env.render(view_mode=RenderView.FULL_SIM_STATE, draw_traces=True, env_id=0)

Training-time evaluation

Rendering during training is controlled by the [eval] section of drive.ini. See that file for available options (human_replay_eval, self_play_eval, eval_interval, etc.).

Sharp edges

• Raylib is not thread-safe. If you create two separate render envs, always call env1.close() before calling env2.render().
• Headless mode derives window dimensions from map bounds automatically; no manual resolution configuration is needed.