Test your robot. See where it fails.

Choose a robot and model in the arcade, try a challenge and replay what happened. Local practice uses a scripted arm and does not score a named policy.

Plan 10 to 100,000 total task attempts, with visible runtime pricing. The default illustration is US$171.43 for 100,000 episodes, based on 500 episodes per worker-hour, US$0.60 per worker-hour and a 30% gross margin. These are assumed inputs, not measured policy throughput or a quote.

The first execution kit covers ten Panda + π₀.₅ / LIBERO Spatial episodes. Larger counts are budget forecasts. Startup, storage, transfers, retries, external inference, operator support and taxes are excluded from the runtime price example.

Hosted policy tests are not yet available. The first real-policy GPU run still needs verification. Larger packs follow measured runtime and cost.