SELF-PACED WORKSHOP

Troubleshooting

Check environment, files, and arguments first. Preserve failed outputs for comparison.

Quick reference

Start with the environment report

Run the command below in a terminal at the repository root. If it specifies an output directory, use a new one to retain earlier results.

python3 scripts/workshop.py doctor
python3 scripts/workshop.py verify-models

The first failure is more useful than repeatedly launching a full training run. Check the repository directory, the selected runtime, and all model checks.

Common issues

Symptom	What to check
No GUI window	Use the GPU desktop terminal and check DISPLAY and the driver. A remote text shell is not a graphical desktop.
First launch is slow	Check for downloads and compilation. Allow initialization and avoid starting duplicate instances.
Output directory exists	Choose a new run name and preserve the previous result.
Out of memory	Reduce training environments or detector batch size. Avoid concurrent simulations.
Model hash mismatch	Restore the repository model and check for accidental overwrites.
Practice policy walks poorly	A short run teaches the workflow. Inspect reward, episode length, and evaluation. Continue with fixed models.
Robot stops or does not score	Inspect the task report and observation validity. Stopping can be appropriate when information is missing; it is not a goal.

Inspect a run

Run the command below in a terminal at the repository root. If it specifies an output directory, use a new one to retain earlier results.

python3 scripts/workshop.py report --run outputs/football-01

When reporting an issue, include the command, seed, environment report, and first error. A success/failure label alone is not enough. Do not include passwords or access tokens.