Sequential Decision Lab / measured simulations
Watch the policies. Inspect the evidence.
Official MiniGrid navigation and MuJoCo locomotion, compared across independently fitted policies. Every animation uses the same first held-out reset. Failures remain visible alongside successful episodes.
Learning curves
Official validation return. Each line is one independently fitted policy; rewards are unshaped during evaluation.
Compare all presets on this environment
Training SD is across three fitted policies. Each policy's test mean uses 100 fixed resets. Planning and random references have no fitting seeds.
| Preset | Fits | Mean test return | Training SD | Mean goal success | Access |
|---|